# Connect a custom MCP server or API Source: https://guides.runner.now/connections/connect-your-own-mcp Connect a public MCP server or save an API key for a custom API. Use **Custom MCP** when a service exposes a public MCP server. Use **Custom API** when it gives you an API key and an HTTP API instead. Add a custom connection only when a specific job needs it, then start with the narrowest tool permissions that work. Both options live in **Connected Apps**. Looking for Gmail, Calendar, Slack, or another built-in option? Start with [Connect your apps](/get-started/connecting-accounts). **Never paste passwords, API keys, bearer tokens, or reusable secrets into a session.** Enter them only in the matching credential field under **Connected Apps**. ## Connect a custom MCP server Your MCP server needs a public HTTPS endpoint that Runner can reach. Use the full endpoint, including a path such as `/mcp` or `/sse` when the server requires one. Plain HTTP, private network addresses, query strings, and URL fragments aren't supported. In Runner, click **Connected Apps** in the sidebar. Find **Custom MCP**, then click **Connect**. In **Connect an MCP server**, fill in **Name** and **HTTPS endpoint**. For example: `https://mcp.example.com/mcp` Open **Authentication** and choose one: * **No authentication** for a server that doesn't require credentials. * **Bearer token** to enter a token in the **Bearer token** field. Runner encrypts the token before storing it and doesn't show it again. * **Sign in with the provider** for OAuth. Runner opens the provider's sign-in flow after you connect. Click **Connect server**. For OAuth, finish signing in with the provider. Runner discovers the tools advertised by the server. The connection then appears under **Connected**, with each tool listed in its **Tools** section. Leave the advanced OAuth settings alone unless the server owner gave you a pre-registered OAuth client. In that case, click **Show Advanced** and enter its **Client ID** and optional **Client secret**. If the provider asks for a callback URL, use **Copy** beside **Callback URL** and register it there. ## Connect a custom API No MCP server? If the service has an HTTP API and gives you an API key, add a **Custom API** connection instead. In **Connected Apps**, find **Custom API**, then click **Connect**. In **Connect an API key**, fill in: * **Name** * **API key** * **Docs or API URL (optional)**, using a public HTTPS documentation page or API base URL Click **Connect API key**. Runner encrypts the key before storing it. The key doesn't appear in your session and isn't displayed again. Runner won't use the stored key until you approve where it can be sent. If Runner finds a possible destination from public API documentation or the URL you entered, open the connection, review **Suggested API URL**, then click **Approve API URL**. Discovery reads only public pages and never sends your key. If there's no suggestion, find **Approved API URL**, enter the public HTTPS base URL in **API URL**, then click **Save URL**. Once you approve the URL, the connection gets two tools: * `read` makes `GET`, `HEAD`, and `OPTIONS` requests to the approved API URL. * `request` makes any request to the approved API URL, including writes. ### What approving an API URL does Runner approves the exact HTTPS scheme and host, not a path. If you enter `https://api.example.com/v1`, the approved URL is `https://api.example.com`. Any path, query string, or fragment is removed. Runner sends the stored key only to that approved scheme and host. A redirect never receives the key. If the suggested API host differs from the documentation host, Runner shows a warning so you can review the destination before approving it. ## Set tool permissions Open either custom connection and find **Tools**. Each tool shows the settings it supports, so you may not see all four on every tool. | Setting | What it does | | --------- | ------------------------------------------------------------------ | | **On** | Lets Runner read with the tool. | | **Ask** | Runner stages an **Action Card** and waits for your approval. | | **Allow** | Runner acts without asking, where supported. | | **Off** | Blocks the tool unless a user-created Runner has its own override. | Start action tools on **Ask**. Move a tool to **Allow** only when you want Runner to use it without approval. ## Manage or remove a custom connection Open the connection under **Connected** to manage it: * For an MCP server, use **Refresh tools**, **Reconnect** for OAuth, or **Remove** and then **Remove server**. For bearer authentication, enter a new value under **Bearer token**, then click **Replace token**. * For a Custom API connection, use **Approved API URL** and **Save URL** to change the destination, **Replace API key** and **Replace key** to store a new key, or **Remove** and then **Remove connection** to delete it. Removing a connection erases its stored credential from Runner. Revoke the token or key with the provider too. # Tool permissions Source: https://guides.runner.now/connections/tool-permissions Decide, tool by tool, what Runner can read, change, or send. Tool permissions control what Runner can read, change, or send on each connected account. Every connected app is a bundle of tools. Gmail has "search inbox" and "send email." Those tools have separate settings, so Runner can read while every send waits for review. ## The four settings Open **Connected Apps**, select an account under **Connected**, and find **Tools**. | Setting | What it means | | --------- | ------------------------------------------------------------------------------------------------ | | **On** | Runner can read with this tool. Reading tools only. | | **Allow** | Runner performs this action without asking. | | **Ask** | Runner stages an **Action Card** for the action and waits for your approval. | | **Off** | Runner can't use this tool with this account, unless you've allowed it in a user-created Runner. | **Staging an Action Card does not perform the action.** Review it, then approve or reject it. A pending **Action Card** expires 24 hours after Runner stages it. If it expires, the action hasn't happened and the card can no longer be approved. Ask Runner to stage it again if you still want it. *** ## Settings are per account A tool setting applies to the connected account you're viewing. If you connect two Gmail accounts, sending can be **Ask** for work and **Allow** for personal. Changing one doesn't change the other. *** ## A setup that works Why not allow everything? Because **Allow** skips review. Start with this: 1. **Start with reads on and writes on Ask.** Runner can read, while proposed writes wait for your approval. 2. **Move routine actions to Allow when you're ready.** Runner will perform those actions without an **Action Card**. 3. **Keep sending and hard-to-undo changes on Ask.** Email sends, Slack posts, and deal updates wait as **Action Cards**. 4. **Use Off when the tool shouldn't run.** A user-created Runner can still override **Off** for its own runs. Keep payments, legal submissions, relationship-sensitive messages, and hard-to-reverse changes on **Ask** or **Off**. *** ## Runner-specific overrides Your **Connected Apps** setting is the baseline. On a user-created Runner, open **Tool permissions** to add an override. Anything you don't override stays **Inherited** and uses the **Connected Apps** setting. The Runner override takes priority during that Runner's runs, including over **Off**. It applies to that tool across the connected accounts the Runner uses. For example: * An inbox-triage Runner can use **Allow** for archiving while the connected account stays on **Ask** elsewhere. * A research Runner can set its write tools to **Off** during its runs. * A user-created Runner can use a tool that's **Off** outside that Runner. Built-in Runners are the exception. **Heartbeat** and **Daily Brief** don't have per-Runner overrides. They use the tool settings in **Connected Apps**. *** ## What's next? See how each Runner gets a saved job, schedule, and tool settings. Bring custom tools into Runner, with the same controls. # Connect your apps Source: https://guides.runner.now/get-started/connecting-accounts Connect the apps where the work lives, then choose what each tool can do. Connect the apps where the work lives, then choose what Runner can read, change, or send. Don't connect everything because you can. Start with one job and the sources it needs. } title="Gmail" /> } title="Calendar" /> } title="Slack" /> } title="Linear" /> } title="Notion" /> } title="Google Workspace" /> } title="HubSpot" /> } title="Salesforce" /> } title="Mixpanel" /> } title="GitHub" /> ## Connect the tools for the first job | AI employee | Start with | | ------------------------- | ------------------------------------------------------- | | **Personal Assistant** | Gmail and Google Calendar | | **Pipeline Inspector** | HubSpot or Salesforce, plus Gmail | | **Invoice Check** | Gmail or Drive, Google Sheets, and the agreement folder | | **Customer-impact check** | Mixpanel, Linear, and GitHub | Add more only when the job needs them. ## How it works When you connect an app, Runner opens the provider's sign-in and permission screen. Review the access it requests, choose an account if prompted, and approve it. You can then set a permission for each tool on that account. Security and policy details change. The current versions live on Runner's [Security](https://runner.now/security), [Privacy Policy](https://runner.now/privacy), and [Terms of Service](https://runner.now/terms) pages. *** ## Connect your first app In Runner, click **Connected Apps** in the sidebar. Search by name or find the app under **Available**, then click **Connect**. If this is your first connection, review the notice and click **Connect**. Finish the provider's sign-in steps in the window that opens. Once the account appears under **Connected**, open it and review **Tools**. Start read tools on **On** and write tools on **Ask**. Ask for something small: * *"What's on my calendar today?"* * *"Show me my unread emails"* * *"Draft a message to #general on Slack. Don't send it yet."* *** ## What you can connect }> Read and search your inbox, draft replies, summarize threads, flag messages that need a response. }> Show today's meetings, create events, find open time slots, warn about scheduling conflicts. Gmail and Google Calendar are separate connections. Connect both if you want Runner to work across your inbox and calendar. }> Read channels and DMs, send messages on your behalf, summarize what you missed, find specific conversations. }> Show assigned tasks, create issues from conversations, update status, summarize what shipped this week. }> Issues, PRs, repos, and code search. }> Search docs, pull relevant info into conversations, create and update pages. }> Search and read Drive files, create folders, and create or edit Docs, Sheets, and Slides. }> CRM updates, deal tracking, contact research. }> Pipeline data, account management, activity logging. }> Product analytics, user funnels, event data. And many more: Calendly, Dropbox, Asana, Intercom, Shopify, Typeform, WhatsApp, and others. The list keeps growing. Missing one? Hit **Request a connector** and tell us. *** ## Work and personal, both connected Have two Gmail accounts? Open **Connected Apps**, find Gmail under **Available**, and click **Add account**. Runner lists each account separately, and you can tell it which one to use: *"Reply from my work email."* To keep the work organized, use [Projects](/get-started/projects). A Project groups related sessions and Runners and gives them shared context. *** ## You stay in control Every connected account exposes its tools, and you decide what each one can do: **On**, **Allow**, **Ask**, or **Off**. Set email search to **On** and sending to **Ask**, for example. Runner can search freely. A send waits in an **Action Card** until you approve it. [How tool permissions work →](/connections/tool-permissions) *** ## Public share links When you choose **Share online** for a session or file, Runner creates a public link. Anyone with the link can view it until you choose **Stop sharing**. Open that session or file's share menu and click **Stop sharing** when you no longer want the link to work. *** ## Reconnect or disconnect **If a connection needs attention**, click **Reconnect** beside the account and finish the provider's sign-in steps. **To remove an account**, open its actions in **Connected Apps** and click **Disconnect**. After Runner confirms the disconnect, that account is no longer offered to Runner. If Runner shows **Access may still be active**, you must also revoke Runner in the provider's account or security settings. *** ## What's next? Put one valuable job on a schedule with the tools it needs. Connect a public MCP server or save an API key. # Projects Source: https://guides.runner.now/get-started/projects Group related sessions and Runners, then give them shared context. Projects group related sessions and Runners and give them shared context. Put the sessions and Runners for Sales in one Project. Do the same for a client, a hiring push, a house move, or the newsletter. If you used workspaces or labels on Runner Desktop, Projects replace both and add shared context. ## What a Project holds Each Project groups three things: * **Sessions.** Every conversation related to this work lives under the Project in your sidebar. * **Runners.** Your "Pipeline Inspector" can live inside the Sales Project, next to the sessions it creates. * **Context.** This is the difference. Keep reading. *** ## Project context Why are Projects better than folders? One field: **Project context**. Open any Project, click **Edit**, and add the background every session and Runner in that Project should know: who the client is, what the goal is, what "done" means, and links that matter. Write it once and you won't need to paste the same background into every new session. Useful Project context names the goal, definitions, people, and source locations that matter. > **Project context for "Q3 Fundraise":** We're raising a \$5M seed. Target close is October 15th. Lead conversations are with Haun and Founders Fund. Kent owns investor communication. The data room lives in Google Drive under /Fundraise. Project context gives sessions and Runners shared background. It doesn't replace a Runner's saved prompt or tool permissions. Put job rules in the prompt, then enforce hard action boundaries with **Ask** or **Off**. A Project doesn't make Runners coordinate or pass results to one another. Keep live state in a source app each Runner can read. *** ## Create a Project In the sidebar, click **New project** and give it a short, clear name. "Q3 Fundraise". "House Move". "Newsletter". Open the Project and edit its details. Fill in the **Project context** field with the background that applies to everything inside it. Use the **Start a new session** box right on the Project page. Or drag existing sessions and Runners in from the sidebar. *** ## Good to know * **Sessions and Runners can move.** Drag them between Projects in the sidebar. A Runner that doesn't belong anywhere just lives outside of Projects, which is fine. * **Recent catches strays.** Sessions you haven't filed show up in the **Recent** section of your sidebar. * **Projects sync everywhere.** Same Projects on your laptop, phone, and iPad. [Add Runner to your home screen →](/get-started/runner-on-your-phone) *** ## What's next? Put one recurring job inside the Project that holds its context. Project context belongs to one Project. Personal Memory can be used across sessions and Runners. # Runner on your phone Source: https://guides.runner.now/get-started/runner-on-your-phone Add Runner to your home screen and check results from anywhere. Getting Runner onto every device was one of the main reasons we moved to the web. There's no app store download. You add [app.runner.now](https://app.runner.now) to your home screen and it behaves like a native app: its own icon, full screen, one tap away. Same sessions, same files, same memories as your laptop. Dictate a task while you're walking to a meeting. Check the result or approve an Action Card when you sit down. ## iPhone and iPad Go to [app.runner.now](https://app.runner.now) in Safari or Chrome and sign in. Everything you set up on your laptop is already there. Typing app.runner.now into the browser on iPhone Tap the **Share** button (the square with the arrow) in the address bar. Runner open in Safari with the Share button highlighted Scroll down the share menu until you see **Add to Home Screen**. If you don't see it, tap **View More** first. The share menu with Add to Home Screen highlighted Make sure **Open as Web App** is toggled on, then tap **Add**. Runner lands on your home screen like any other app. The Add to Home Screen dialog with Open as Web App toggled on ## Android Go to [app.runner.now](https://app.runner.now) and sign in. Tap the **three-dot menu** in the top right corner. Tap **Add to Home screen** (on some phones it says **Install app**), then confirm. This works on anything with a browser: Chromebooks, tablets, your partner's iPad. Sign in, add to home screen, done. *** ## What people use it for * **Voice notes on the go.** Tap the mic, describe what you need, board your flight. Runner works while you're in the air. * **Reviewing Action Cards.** When a write tool is set to **Ask**, Runner stages an **Action Card** you can review from your phone. **Allow** performs the action without a card. **Off** blocks the tool unless a user-created Runner has its own override. * **Reading your Daily Brief.** Once you turn its schedule on, each 8am run is available on your phone. *** ## What's next? Haven't built your Personal Assistant yet? Start here. See the recurring jobs and results you'll check from your phone. # Runner lets you build a fleet of AI employees for your highest-value jobs Source: https://guides.runner.now/index Each Runner owns one clear job and works on a schedule across your connected apps. **Still on the Runner desktop app?** Runner v1 retires September 1st, 2026. Moving to the web takes about 15 minutes and brings your memories, workflows, and automations with you. [Migrate now →](/legacy/runner-desktop) Each Runner is an AI employee with one clear job. One follows proposals every morning. Another checks vendor invoices every Friday. Another gets your day in order at 9am. They handle repeatable work across your connected apps and the cloud browser. You stay focused on judgment, relationships, creative decisions, and the work only you can do. Use a session to brief a Runner, check a result, or improve the saved prompt. The scheduled work keeps running after you close the browser. *** ## Get started Build a Personal Assistant that checks Gmail and Calendar every morning. Find the jobs tied to revenue, cost, time, or customer delivery. Connect Gmail, Calendar, Slack, and more, then decide what each tool can do. Use Projects to keep related sessions, Runners, and background together. *** ## Build the fleet one job at a time Start with one recurring job you can measure. Get it working, then add another Runner. **Write the job.** A Runner's saved prompt says what to check, how to decide, when to stop, and what a useful result contains. **Connect the sources.** Connected apps and the cloud browser are how the work gets done. You connect the accounts, set tool permissions, and name the sources each job should use. **Set the schedule.** Every weekday at 8am. Friday at 4pm. The first day of each month. Each run starts in a separate session and keeps going in the cloud after you close the browser. **Set the review point.** **On** lets Runner read. **Ask** stages an **Action Card** for review. **Allow** acts without asking. **Off** blocks the tool unless a user-created Runner has its own override. **Make the job last.** Runner keeps the saved prompt, schedule, Project, intelligence setting, and any Runner-specific tool permission overrides. Previous results don't carry automatically. Put rules for every run in the saved prompt. Use Project context for background. Each Runner owns its own job. Runners don't automatically hand work to one another or receive another Runner's output. You decide how their jobs fit together. AI employees with saved jobs, schedules, sources, rules, and tool permissions. See recurring jobs for revenue, cost, time, and delivery. Reusable instructions you start with `/slug` whenever you need them. *** ## Which AI employee would make the biggest difference? The best first AI employee sits close to money or a customer commitment. Add proposal follow-up, client onboarding, and delivery support. Add an executive briefer, revenue follow-through, and customer risk watch. Add lead response, pipeline inspection, and proposal follow-up. Add invoice checks, vendor renewal watch, and report assembly. [See AI employees by role →](/use-cases/overview) *** ## Get more out of Runner Keep writes on **Ask**, check early work, and fix the saved job. Save the facts and preferences Runner can use across sessions and Runners. Share Runner with people you work with. *** ## Stay in the loop Latest product updates as they ship. Join the community or reach out to the team. Learn more about Runner. # Moving from Runner Desktop Source: https://guides.runner.now/legacy/runner-desktop Runner Desktop (v1) retires September 1st, 2026. Move to Runner on the web in about 15 minutes. Runner started as a desktop app. We launched in April, you gave us a ton of feedback, and the loudest theme was simple: I want Runner everywhere, and I don't want it to stop when my laptop closes. Runner v2 is that. It lives at [app.runner.now](https://app.runner.now), runs on a dedicated virtual computer in the cloud, and works from any browser on any device. This page gets you moved over. **The dates that matter:** * **August 3rd, 2026:** Last new features shipped to Runner Desktop. * **August 14th, 2026:** Migrate before this date and we'll give you a month of Runner free. Message [charlie@runner.now](mailto:charlie@runner.now) on Slack or email to claim it. * **September 1st, 2026:** Support for Runner Desktop ends. After this date you can only use it with your own API key. ## What moves, what doesn't The migration brings over the things Runner has learned about you: | From Desktop | On the web | | ------------------- | --------------------------------------------------------------------- | | Memories | Move over as-is | | Skills | Appear under **Workflows** | | Automations | Appear under **Runners**, turned off by default so nothing runs twice | | Individual chats | Moved one at a time with `/move-to-runner-web` | | Billing and credits | Already linked. Accounts are tied by email address. | | Connected apps | **Don't move.** For security reasons you reconnect them on the web. | *** ## Migrate in five steps Open any browser and go to [app.runner.now](https://app.runner.now). Use the same email address you use on desktop. Accounts are tied together by email, so your billing and credits are already there waiting for you. Open Runner Desktop, click the settings gear in the bottom left, and under **General** hit **Check Now** next to "Check for updates". The migration tools shipped in v1.1.38, so make sure the version says 1.1.38 or later. Runner Desktop settings showing version 1.1.38 and the Check Now button In your desktop **Settings**, click **Sync to Web**, then press the button to start. This copies your memories, skills, and automations to your web account. It takes about 1 to 5 minutes depending on how many files you have. Automations land under **Runners** turned off by default. Skills land under **Workflows**. The Sync to Web page in Runner Desktop settings It's a one-way copy. Run it again anytime (the button reads **Resync** after the first run) to refresh the web with your latest desktop state. For any individual chat session you want to keep, open that chat on desktop and type: ```text theme={null} /move-to-runner-web ``` The **Move to Runner Web** skill pops up as soon as you start typing `/move`. Typing /move in a desktop chat surfaces the Move to Runner Web skill Runner summarizes the chat's full context and copies it for you. Open a new session on the web and paste (`Cmd+V` or `Ctrl+V`). The web session picks up right where the desktop one left off. Runner confirming the session is copied and ready to paste into Runner web You don't need to do this for every chat. Just the ones with context worth keeping. Go to **Connected Apps** on the web and reconnect Gmail, Calendar, Slack, and the rest. Why don't connections migrate? Security. Your app credentials never leave the environment they were created in. Reconnecting takes about a minute per app. That's it. You're on Runner v2. ## Review migrated Runners Migrated automations arrive paused. Before you turn one back on: 1. Give it one clear job tied to revenue, cost, time, or customer delivery. 2. Check its saved prompt, schedule, timezone, Project, model, and permissions. 3. Click **Run now** and inspect the result and tool activity. 4. Turn on one schedule at a time so you can see what each Runner adds. If two migrated Runners do the same job, keep the better one. A few clear jobs beat a crowded sidebar. *** ## Uninstall the desktop app We strongly recommend deleting Runner Desktop once you've migrated. It won't be maintained past September 1st, and nothing on the web syncs back to it. * **Mac:** Open your Applications folder, right-click Runner, and choose **Move to Trash**. * **Windows:** Start menu, then **Uninstall Programs & Applications**, and remove Runner. Right-clicking Runner in the Mac Applications folder and choosing Move to Trash *** ## Desktop FAQ We'll never force you to delete it. But we heavily recommend moving on. Desktop and web are two separate apps that don't sync with each other, desktop stopped getting new features on August 3rd, and support ends September 1st, 2026. After that, desktop only works with your own API key. New web-first Runner features ship to the web app. Yes. Accounts are linked by email address. Sign in on the web with the same email you use on desktop and your plan and credits are already there. Migrated automations arrive under **Runners** turned off by default, so nothing fires twice while you're moving over. Open the **Runners** tab and enable the ones you want. They now run 24/7 in the cloud, laptop open or not. Skills are now **Workflows**: reusable instructions you invoke with `/slug` in any message. Workspaces are replaced by **Projects**, which organize your sessions and Runners by what you're working on. [More on Projects →](/get-started/projects) Yes, happily. Message [kent@runner.now](mailto:kent@runner.now) or [charlie@runner.now](mailto:charlie@runner.now) on Slack or email and we'll set up a call. We'll do the migration with you and show you how to get the most out of Runner while we're at it. *** ## What's next? Build your Personal Assistant on the web in about 15 minutes. The other big reason we went web-first. Add Runner to your home screen. # Quickstart Source: https://guides.runner.now/quickstart Build your first AI employee: a Personal Assistant for your 9am attention brief. Build your first AI employee: a Personal Assistant for your 9am attention brief. It reads Gmail and Google Calendar, spots what needs attention, and gives you a plan. You set the priorities, review sensitive replies, and make final decisions. | Runner | Sources | Result | You decide | | ---------------------- | ------------------------- | --------------------------------- | -------------------------------------------------- | | **Personal Assistant** | Gmail and Google Calendar | A short plan every morning at 9am | Priorities, sensitive replies, and final decisions | You'll connect two apps, check their permissions, create the Runner, then run it once. The screenshots show mine, named Shiela: Kent's Chief of Staff. That's the display name. In this guide, her only job is the 9am attention brief. Give yours any name you want. Go to [Runner](https://app.runner.now/app) in any browser and sign in. There's nothing to install. Open **Connected Apps** and connect Gmail and Google Calendar. Open **Tools** on each account. Set read tools to **On** and write tools to **Ask**. **On** lets Runner read. **Ask** stages an **Action Card** for review. **Allow** acts without asking. **Off** blocks the tool unless a user-created Runner has its own override. [Connect your apps →](/get-started/connecting-accounts) Open **Runners** in the sidebar. You'll see **Heartbeat** and **Daily Brief** there too. Both ship paused, so leave them that way for now. Click **New runner**. The New runner button on the Runners page Choose **Ask Runner to create it**. This opens a new session with the first line already filled in. An arrow points to Ask Runner to create it in the Create a runner window Set the session intelligence to **Normal** before you continue. Add this below the prefilled line: ```text theme={null} Create a Personal Assistant Runner for me. Every day at 9:00am in my timezone, review my connected Gmail and Google Calendar. Give me a short plan for the day with: 1. Three items that may deserve my attention, with the source and why. 2. Meetings that need preparation. 3. Commitments and open loops that need attention. 4. Schedule conflicts, deadlines, or important messages I may have missed. Be specific. If the source is unclear, say so instead of guessing. If a write action would help, stage it as an Action Card and wait for my approval. Don't add Runner-specific tool permission overrides. My write tools are set to Ask in Connected Apps. Ask me whatever you need. Before you create the Runner, show me the final name, prompt, schedule, and timezone. Wait for my confirmation. ``` Click **Send**, then answer Runner's questions. Give it real details. One current priority and one thing it shouldn't touch will do more than another paragraph about being proactive. Read the final name and prompt. Check the 9am schedule and timezone. Because you started from the main **Runners** page, it begins under **No project** with no Runner-specific tool permission overrides. **A user-created Runner becomes active as soon as it's created.** Confirm only when you're ready for the schedule to start. Open **Runners** and select your new Personal Assistant. Shiela: Kent's Chief of Staff showing the Status and Action Cards controls Treat the saved prompt as the job description. Check the prompt, schedule, timezone, **Project**, and **Runner model**. Choose a Project if you want its context. Keep the model on **Normal** for now. Scheduled Runners don't use **Auto**. **Ultra** is for harder jobs and uses more of your plan. Under **Tool permissions**, leave **No overrides** while you test. The Runner uses your current **Connected Apps** settings. If you add an override later, it takes priority, including over **Off**. The **Action Cards** switch carries unresolved cards into later runs. It doesn't carry old run output, and a pending card still expires after 24 hours. Pick an avatar. It saves immediately. If you changed anything else, click **Save changes**. The avatar picker and Save changes button on a Runner Don't wait until tomorrow morning. Click **Run now**. An arrow points to Run now on Shiela: Kent's Chief of Staff The test creates a separate session and keeps you on the Runner detail page. You can run a paused Runner this way too. Open **Recent runs** and click the run's timestamp. You can also click your Personal Assistant's existing row under **Recent** in the sidebar. An arrow points to Shiela: Kent's Chief of Staff under Recent Talk to it the way you would a colleague. Be direct: * *"I didn't like that example. Be more specific."* * *"I don't have a car. Use transit options next time."* * *"Tell me about key events at least two days before they happen, not the morning of."* If the feedback should apply every morning, add it to the saved **Prompt**. Put durable facts and preferences in [Personal Memory](/using-runner/memory). ## What carries into the next run Each run starts in a separate session. Your saved prompt, schedule, timezone, Project, and intelligence setting stay with the Runner. So do any Runner-specific tool permission overrides. Tools without an override use your current **Connected Apps** settings. The previous run's output doesn't come along automatically. Put standing instructions in the saved prompt, background in Project context, and durable personal details in Personal Memory. *** ## What's next? Find recurring work with a result you can measure. Decide what waits for review and what can run without asking. Save the facts and preferences Runner should know across sessions. # Create your own Runner Source: https://guides.runner.now/runners/create-your-own Add one AI employee with a valuable job, schedule, sources, rules, and tool permissions. You're adding one AI employee, not asking a generic assistant to "help with work." Give the Runner one job. Make sure that job protects revenue, removes cost, gives back meaningful time, or keeps a customer promise from slipping. ## Write the job description first | Part | What to decide | | ------------------ | ------------------------------------------------------------------------------- | | **Job** | The one result this Runner should produce | | **Value** | The time, cost, revenue, or customer promise attached to it | | **Sources** | The connected apps, files, records, or sites it should use | | **Schedule** | When the job runs and which timezone it follows | | **Stop condition** | When Runner should report that no action is needed and stop | | **Review** | What can happen directly and what must wait for an Action Card | | **Human work** | The judgment, relationship, creative choice, or final decision that stays yours | | **Score** | The number you'll check after the first few runs | There are two ways to add the Runner: 1. Fill in the form yourself. 2. Ask Runner to help you design the job. Use the second one when the job is still fuzzy. Use the form when you already know the brief and schedule. **A new Runner is active as soon as it's created.** Check every field the creation path shows before you confirm. Then open the new Runner and review its Project, model, and tool permissions before the first scheduled time arrives. ## Ask Runner to create it with you Open **Runners**, click **New runner**, then choose **Ask Runner to create it**. Set the new session to **Normal** before you continue. Say what should happen, which sources to use, when it should run, what needs review, and what a useful result looks like. Add this line when you want a review before creation: ```text theme={null} Ask me whatever you need. Show me the final name, prompt, schedule, and timezone. Wait for my confirmation before you create the Runner. ``` Read the proposed name, prompt, schedule, and timezone. Starting from the main **Runners** page creates it under **No project** with no Runner-specific tool permission overrides. Confirm only when you're ready for the Runner to become active. Open the new Runner. Choose a **Project** if it needs shared background, verify **Runner model**, and review **Tool permissions** before the first scheduled run. ## Fill in the job yourself Open **Runners**, click **New runner**, then choose **Create manually**. Fill in: * **Name** * **Prompt** * **Schedule** and **Timezone** * **Project**, when the job belongs with other work * **Runner model**, using **Normal** or **Ultra** Click **Create runner** only after you've checked the schedule. The new Runner starts active. ## Set tool permissions For connected-app writes: * **Ask** stages an **Action Card** for review. * **Allow** performs the action without asking. * **Off** blocks the tool unless this Runner has its own override. A new Runner starts with **No overrides** and uses your current **Connected Apps** settings. Add a Runner-specific override only when this one job needs a different rule. That override takes priority, including over **Off**. Start writes on **Ask** when you're testing a new Runner. Move a specific tool to **Allow** only when you want that Runner to use it without stopping for review. [How tool permissions work →](/connections/tool-permissions) ## Choose what carries into each run Each run starts in a separate session. Previous run output is not automatically supplied to the next one. Use Personal Memory for durable personal facts. Use Project context for background. Keep the job's instructions, thresholds, and review rules in the saved prompt. The **Carry unresolved Action Cards into new runs** switch does one narrower job. It lets an unresolved **Action Card** appear in a later run. A pending **Action Card** still expires 24 hours after it was staged. ## Check the first three runs Open the Runner and click **Run now**. The test creates a separate session and keeps you on the Runner detail page, even if the Runner is paused. Open **Recent runs** and click the run's timestamp. Check the result, tool activity, and any **Action Cards**. Review the first three runs this way. Fix job instructions in the saved prompt and incorrect background in Project context. Then leave the schedule active or pause it while you adjust the job. Add another Runner after the first three runs are useful and consistent. *** ## What's next? See what carries into each run and when actions wait for review. Find the next job tied to revenue, cost, time, or customer delivery. # Heartbeat and Daily Brief Source: https://guides.runner.now/runners/heartbeat-and-daily-brief Set up the two built-in Runners: a morning brief and a twice-daily check for messages that need a reply. Runner ships with two built-in Runners. Both start paused. | Runner | Job | What waits for you | | --------------- | -------------------------------------------------------- | ------------------------------------------------ | | **Daily Brief** | Your morning attention brief | The priorities and decisions that need you today | | **Heartbeat** | Checks twice each weekday for messages that need a reply | Replies and decisions that still need a person | Test them one at a time. It's easier to see what each one adds, and you won't create three versions of the same morning update. If you already built a Personal Assistant, check for overlap before turning on either schedule. ## Daily Brief Daily Brief is scheduled for 8am every day in your timezone. It reads your calendar, inbox, and supported connected apps, then writes one update about what needs you today. The stock prompt only reads from your connected apps. Its result appears in a new session, and it asks Runner to email you when it's ready. Email delivery depends on **Settings**, then **Notifications**, including the daily limit. To test it: 1. Open **Runners**, then **Daily Brief**. 2. Review the prompt, schedule, and timezone. 3. Click **Run now**. 4. Read the result in the new session. 5. Turn on **Status** only if you want the daily schedule. Change the prompt when you want a different brief. Be specific: *"Lead with anything from investors. Skip newsletter summaries."* ## Heartbeat Heartbeat runs at 7am and 4pm on weekdays in your timezone. It checks connected communication channels and sorts the result into: * **Needs you.** Something is waiting on a decision or reply. * **Drafted.** A proposed reply is ready for review. * **Heads-up.** Worth knowing, but no action is needed yet. * **Could not check.** Runner couldn't reach a source. When the relevant write tools are on **Ask**, the stock prompt can stage up to three **Action Cards** per run. If nothing needs you, it gives you one line and stops. ## Permissions for built-in Runners Heartbeat and Daily Brief don't have their own **Tool permissions** section. They use the settings on each account under **Connected Apps**. If you want every outward write to wait for review, keep those tools on **Ask**. **Ask** stages an **Action Card**, **Allow** acts without asking, and **Off** blocks the tool. [Review tool permissions →](/connections/tool-permissions) ## What carries into the next run Each built-in run starts in a separate session. It can use Personal Memory, but it doesn't automatically receive the previous run's output. If a durable preference matters tomorrow, put it in Personal Memory or the saved Runner prompt. *** ## What's next? Build a Personal Assistant that checks in every morning. Give one recurring job a schedule, rules, and tool permissions. # Runners Source: https://guides.runner.now/runners/overview A Runner is one AI employee with a recurring job, schedule, sources, rules, and tool permissions. A Runner is one AI employee assigned one recurring job. A Runner is a saved prompt with a schedule. Tell it what to check, which sources and rules to use, and what the result should contain. Then choose when it runs and close your laptop. Every weekday at 8am. Friday at 4pm. The first day of each month. Start with a job that costs you something when it gets missed: a proposal waiting on a reply, an invoice exception, a customer commitment, or a report you rebuild every Friday. ## Give every Runner a clear brief | Part | The question to answer | | -------------- | ------------------------------------------------------------------------------- | | **Job** | What should this Runner do every time? | | **Value** | Does it protect revenue, remove cost, save time, or protect a customer promise? | | **Tools** | Which connected apps, files, or sites does the job need? | | **Handoff** | What can happen directly, and what should come back to you? | | **Human work** | Which judgment, relationship, creative choice, or final decision stays yours? | Each Runner works independently. Runners don't automatically hand work to one another or receive another Runner's output. You decide how their jobs fit together. ## What carries into a run A Runner keeps its saved prompt, schedule, timezone, Project, intelligence setting, and any Runner-specific tool permission overrides. Tools without an override use your current **Connected Apps** settings. Each run itself starts in a separate session. It can use: * **Project context** as background, when the Runner belongs to a Project * **Personal Memory**, for durable facts and preferences saved there * **Unresolved Action Cards**, when carry is turned on and the cards haven't expired It does **not** automatically receive the previous run's output. If a result needs to matter next time, put a durable personal fact in Personal Memory or a standing job instruction in the saved prompt. Project context can hold shared background. ## When an action waits for review The tool setting decides: * **On** lets Runner read with the tool. * **Ask** stages an **Action Card** for review. * **Allow** performs the action without asking. * **Off** blocks the tool, unless the user-created Runner has its own override. A pending **Action Card** expires 24 hours after it's staged. Turning on **Carry unresolved Action Cards into new runs** can show that card in the next run, but it doesn't extend the 24-hour limit. User-created Runners have a **Tool permissions** section for overrides. **Heartbeat** and **Daily Brief** don't. The built-ins use the settings under **Connected Apps**. [How tool permissions work →](/connections/tool-permissions) ## What you set for each Runner Each Runner has: * **Prompt.** The job, sources, rules, and output you want. * **Schedule.** Daily, weekdays, weekly, monthly, or custom. * **Timezone.** The clock the schedule follows. * **Project.** Optional organization and background for the work. * **Intelligence.** **Normal** or **Ultra**. Scheduled Runners don't use **Auto**. * **Action Card carry.** Whether unresolved cards can appear in a later run before they expire. * **Avatar.** A quick way to spot it in the sidebar. ## Check recent runs Open a Runner to see recent runs. Each row shows whether the run was scheduled or manual, when it started, and its status. Click the run's timestamp to open that run's session. That's where you'll find its result, tool activity, **Action Cards**, or failure message. Need a test before the next scheduled time? Click **Run now**. A paused Runner can run once without turning its schedule on. ## Ask for an email when it's done Put the request in the Runner's prompt: *"Email me when this report is ready."* Email delivery also depends on **Settings**, then **Notifications**, including your daily limit. *** ## Built-in Runners Runner ships with two paused Runners: * **Heartbeat** checks communication channels twice on weekdays and surfaces what needs attention. * **Daily Brief** runs every day at 8am and gives you one update across your connected apps. [Set up Heartbeat and Daily Brief →](/runners/heartbeat-and-daily-brief) *** ## What's next? Add one recurring job with its schedule, rules, and tool permissions. Decide what waits for review and what can run without asking. # Best practices Source: https://guides.runner.now/tips/best-practices Build safer Runners with clear jobs, the right sources, and results you can review. Treat each Runner like an AI employee. Start with one narrow job, the sources that job needs, and writes on **Ask**. Writes follow your [tool permissions](/connections/tool-permissions). **Ask** stages an **Action Card** for review. **Allow** acts without asking. **Off** blocks the tool unless a user-created Runner has its own override. ## Start with the tools for one job Don't connect everything on day one. If your first Runner is the Personal Assistant from Quickstart, start with these: } title="Gmail"> Inbox triage, draft replies, follow-up tracking. } title="Google Calendar"> Meeting prep, scheduling, conflict detection. Those two are enough for the morning attention brief. A Pipeline Inspector would start with your CRM and Gmail. An Invoice Check would start with the folder, inbox, and sheet where invoices and agreements live. Connect the next app only when the next job needs it. **Quickstart setup:** Gmail + Google Calendar. That's it. *** ## Group related work in Projects Keep different parts of your life in separate Projects: | Project | What lives there | | ---------------- | ------------------------------------------------------------ | | **Work** | Board prep sessions, your pipeline Runner, the weekly review | | **Personal** | Trip planning, the subscription audit, family logistics | | **Side project** | Every session and Runner for the newsletter | Each Project supplies its context as background to the sessions and Runners inside it. It doesn't replace the instructions in a Runner's saved prompt. Got work and personal Gmail? Connect both in **Connected Apps** with **Add account** and tell Runner which one to use: *"Reply from my work email."* *** ## Start writes on Ask Set the send tool to **Ask** and Runner stages an **Action Card** before anything goes out: 1. Ask Runner to draft the email or message 2. Read it, then make any changes 3. Ask Runner to send it 4. Approve or reject the **Action Card** Later, you can move a specific send tool to **Allow**. That removes the review step, so do it on purpose. *** ## Turn one repeated job into a Runner A one-off request shows you what Runner can do. The bigger payoff comes when you turn repeated work into a clear system. | Too vague | Ready to test | | ------------------------- | ------------------------------------------------------------------------------------------------------------- | | "Help with sales" | "Every weekday at 8am, find open proposals that have waited 3 business days without a reply." | | "Keep an eye on invoices" | "Every Friday, match new vendor invoices in Drive against the active agreements folder and flag differences." | | "Watch product metrics" | "Every weekday at 8:30am, check the three approved funnels in Mixpanel and report changes above 15%." | Name the sources, schedule, rules, what needs your approval, and what the result should contain. Then click **Run now** and read the first result. If you're not sure which job to pick, [start with the work tied to revenue, cost, or time](/tips/tips-and-tricks). *** ## Save preferences that should stick Tell Runner what you want it to remember: *"When you draft emails for me, keep them under 5 sentences. Use a professional but friendly tone. Always end with a clear next step. Remember this."* Then check **Memory**. Once it's saved, future sessions can use it. *** ## Be specific with names and dates Runner works across multiple apps, so precision helps: | Vague | Clear | | ------------------------------ | ------------------------------------------------------------------- | | "Check that email" | "Check the email from Sarah Chen about the Q2 budget" | | "Schedule something next week" | "Find a 30-minute slot on Tuesday or Wednesday afternoon" | | "Update my tasks" | "Mark the API bug task as done and add a note that the fix shipped" | *** ## Review the first three runs Read the first few runs. Fix repeated job mistakes in the saved prompt, not only in one run's session. Correct inaccurate background in Project context. Once a Runner is reliable, review it monthly. Check recent runs, pause work you no longer need, and confirm the schedule and permissions still make sense. Use a simple scorecard: | Check | Ask | | ----------- | ------------------------------------------------------------------------------ | | **Value** | Did it protect revenue, remove cost, save time, or protect a customer promise? | | **Quality** | Was the result accurate, sourced, and useful? | | **Review** | Did the right decisions wait for a person? | | **Cost** | Does the value justify how often it runs and the plan it uses? | | **Need** | Would you notice if the Runner stopped? | If the answers keep being no, fix the prompt or pause the Runner. *** ## Your first week with Runner Here's a good first week. Each step builds on the last. * [ ] [Connect Gmail and Google Calendar](/get-started/connecting-accounts) * [ ] [Build your Personal Assistant](/quickstart) * [ ] Click **Run now** and read the full result * [ ] [Find one repeated job](/tips/tips-and-tricks) tied to revenue, cost, time, or a customer commitment * [ ] Name its sources, schedule, useful result, and what needs your approval * [ ] [Create a Project](/get-started/projects) for shared background * [ ] Put the job rules and review requirements in the saved prompt * [ ] [Create the Runner](/runners/create-your-own) with writes on **Ask** * [ ] Read the first three runs and fix the saved prompt when the same problem appears twice * [ ] Add another Runner only after the first three runs are useful and consistent *** ## What's next? Find the next job tied to revenue, cost, time, or customer delivery. See examples for services, sales, operations, and more. # Refer a friend Source: https://guides.runner.now/tips/referral-program Invite someone to Runner. After they pay, you each get $50 in Runner credits. Share your invite link with someone who's new to Runner. With the standard offer, they get $50 in Runner credits after they pay for Runner. You get $50 too. **Settings**, then **Referrals** shows the current reward and invite limit for your account. The standard limit is 200 referrals. ## Send your invite link Click **Settings**, then **Referrals**. Under **Your invite link**, click **Copy**. Send the full link by text, email, or Slack. They open the link, click **Accept invite**, and sign up for Runner. The link carries your referral code, so there's nothing for them to type during signup. Runner adds \$50 in credits to each account after the referred user pays. Signup alone records a pending referral. It doesn't issue the credits. ## Two time windows matter When someone opens your invite link, Runner remembers it in that browser for up to 90 days. They should finish signup in the same browser so Runner can match the new account to your link. If they open a newer Runner invite link, that link replaces the earlier one. Runner can only attach a referral during the first 24 hours after the account is created. An older account can't claim one. Why two windows? The 90-day window remembers which link they opened. The 24-hour window makes sure the reward goes to a new Runner account. ## Track pending and paid referrals Open **Settings**, then **Referrals**, and look under **Your invites**: * **Invited** means Runner recorded the signup and the referral is pending. * **Converted** means the referred user paid and the referral credits were added. The count above the list shows how many referrals you've used and your current limit. The standard limit is 200. ## Customize your code Your referral code is the last part of your invite link, such as `ada2026`. In **Customize your code**, you can choose a code or generate a new one. Codes must use 4 to 10 lowercase letters and numbers. No spaces or punctuation. **Changing your code disables the old invite link.** Copy the new link before you share again. Referrals already recorded with the old code stay in your history. ## Your accounts stay separate You can see the referred user's signup email, referral dates, and status under **Your invites**. You don't get access to their Runner content. A referral doesn't connect your Runner accounts. You can't see each other's sessions, Runners, connected apps, or other Runner activity. ## Copy-paste messages *"Hey, I've been using Runner for email, calendar, and the follow-ups that usually get lost. If you want to try it, use my link. You'll get \$50 in Runner credits after you pay for a plan: \[your invite link]"* *"Thought you might like Runner. It helps me work through email, calendar, and the repetitive stuff around them. This link gives you \$50 in Runner credits after you pay for a plan: \[your invite link]"* Use the reward amount shown under **Settings**, then **Referrals** if your account has different terms. *** ## What's next? Find a first job tied to revenue, cost, time, or customer delivery. How to get the most out of Runner right away. # Which AI employee should you add first? Source: https://guides.runner.now/tips/tips-and-tricks Find the highest-value jobs for your first AI employees. Think like you're hiring. If you could add one great AI employee tomorrow, which job would you give them? The proposal follow-up nobody owns? The invoice check you rebuild every Friday? The morning plan that keeps your day from getting away from you? Runner lets you build your fleet one job at a time. You still talk to Runner. That's how you describe the job, check the work, and improve the Runner. **Every AI employee needs one clear, valuable job.** Writes follow your [tool permissions](/connections/tool-permissions). **Ask** stages an **Action Card** for review. **Allow** acts without asking. **Off** blocks the tool unless a user-created Runner has its own override. ## Start with a number Look for work tied to one of these: | Value | Ask yourself | | --------------------- | ----------------------------------------------------------- | | **Revenue** | Where do leads, proposals, renewals, or upsells go quiet? | | **Cost** | What spend, invoice, or vendor mistake do we find too late? | | **Time** | What do we rebuild, chase, or check every week? | | **Customer delivery** | What handoff or commitment slips when nobody follows up? | A 45-minute task done five days a week takes 195 hours a year. A forgotten $200 monthly subscription costs $2,400 a year. You don't need perfect math. You need enough of a number to rank this work against everything else. ## Sketch your fleet ### Jobs that protect revenue * Check each weekday for proposals still waiting on a reply. * Check the CRM for new leads with no first response. * Find open deals with no next step or no activity in seven days. * Surface renewals and customer risks before the deadline. * Draft the right follow-up from the email thread, CRM record, and proposal. ### Jobs that cut costs * Match incoming invoices against signed agreements. * Flag renewals, recurring charges, and spend that changed unexpectedly. * Find campaigns spending above your target without producing pipeline. * Catch duplicate work across teams before another person does it again. * Assemble recurring reports without spending another afternoon copying numbers between apps. ### Jobs that keep delivery moving * Turn a signed proposal or order form into an onboarding checklist. * Find projects waiting on a client, owner, or approval. * Draft weekly client updates from the work that actually happened. * Prepare meeting briefs, then draft a decision log from the owners and dates people agreed. * Check that every commitment has an owner and date. Why these? They repeat, they cross several apps, and missing them costs you something. ## Jobs that give time back Personal work is often the easiest place to start. The stakes are lower and the pattern is obvious. * A morning plan built from Gmail and Google Calendar. * A daily check for messages and commitments waiting on you. * A Sunday preview of family events, appointments, and bills. * A monthly subscription and renewal review. * Meeting prep before the first call of the day. Once one of these works, use the same pattern on a business process with a bigger number attached. ## Write the job description 1. **One job.** Not "help with sales." Try "find proposals waiting more than three business days for a reply." 2. **Named sources.** Gmail, HubSpot, a Google Sheet, a Drive folder, or a site in the browser. 3. **A schedule.** Every weekday at 8am. Friday at 4pm. The first day of each month. 4. **Rules and limits.** Who to include, what to ignore, and when to stop. 5. **A result and review point.** A ranked list, a drafted follow-up, an updated record, or an Action Card waiting for approval. 6. **Human work.** The judgment, relationship, creative choice, or final decision that stays with a person. If you can't describe the outcome and the stop condition, the job isn't ready for a Runner yet. More runs use more of your plan. Start daily. Increase the frequency only when the number attached to the work justifies it. ## Ask Runner to help you choose Use this when you need help but don't know which job to start with: ```text theme={null} Help me design the first three AI employees for my business and personal life. Ask me one question at a time about: - leads, proposals, renewals, and revenue - invoices, vendors, subscriptions, and other costs - client delivery and internal handoffs - reports, meetings, and work I repeat every week - personal admin that keeps taking time For each AI employee, give me a name for the job, how often it happens, the apps they need, what they would own, what still needs my approval, which decisions should stay with me, how long the work takes today, and what it costs when I miss it. Rank the best three by revenue at risk, money saved, or time I get back. Don't create anything yet. Show me the three jobs first. ``` This is where chat belongs. It helps you turn a fuzzy problem into a clear job. ## Runner, Workflow, or one-off? * Build a **Runner** when you want an AI employee to own the job on a schedule. * Save a **Workflow** when you want a reusable playbook that you start with `/slug`. * Keep it in one session when the work truly happens once. If a job is tied to real revenue or cost and should happen on a schedule, give it to a Runner. ## Give the first Runner a durable brief Once you've picked the job, give the Runner the details that shouldn't change: ```text theme={null} Create a Runner for [job]. Outcome: [the result that matters] Schedule: [when it should run, including timezone] Sources: [connected apps, folders, records, or sites to check] Rules: [what to include, exclude, and how to make decisions] Review: [which actions must wait for my approval] Human work: [the judgment, relationship, creative choice, or decision that stays with a person] Output: [exactly what a useful result should contain] Stop condition: [when no action is needed] Before you create it, show me the final name, prompt, schedule, and timezone. Wait for my confirmation. ``` A durable Runner keeps its saved prompt, schedule, timezone, Project, intelligence setting, and any Runner-specific permission overrides. Each run still starts in a separate session. Previous run output doesn't carry automatically. Put standing business rules in the saved prompt. Use Project context for shared background. Save durable personal facts and preferences to Personal Memory. Keep live status in the source app, such as your CRM, inbox, or tracking sheet, so the next run can read it again. A user-created Runner becomes active as soon as it's created. Review its schedule and tool permissions before you confirm. ## Add one Runner at a time Don't begin with "run my company." Add one Runner with one job, one schedule, and one clear result. Review the first three runs and fix the saved prompt when the same problem appears again. Once the result is useful and consistent, add the next job. Three useful Runners beat a sidebar full of clever prompts. *** ## What's next? See the jobs that can protect revenue, remove cost, and save time. Give one job a schedule, sources, rules, and tool permissions. Start with a Personal Assistant, Gmail, Calendar, and a useful 9am plan. # Runner for finance and operations teams Source: https://guides.runner.now/use-cases/finance-and-ops Add invoice checks, vendor renewal watch, receivables follow-up, and reporting. An invoice can match last month's total and still be wrong. The rate changed. The contract ended. The same invoice arrived twice. Someone notices three weeks later because the real work is buried across Gmail, Drive, and a tracking sheet. Give that checking job to an AI employee. They bring you the exceptions on schedule. You still approve every payment. Writes follow your [tool permissions](/connections/tool-permissions). Keep messages, records, and approvals on **Ask** while you test. Runner stages each write as an **Action Card**. **Allow** acts without asking. **Off** blocks the tool unless the Runner has its own override. ## Recurring jobs for finance and operations | AI employee | What they own | Why the job matters | You stay focused on | | ------------------------- | ------------------------------------------------------------------------- | ------------------------------------------------- | ------------------------------------------- | | **Invoice check** | Compares new invoices with agreements and the tracking sheet | Finds duplicates, rate changes, and missing terms | Approval and payment | | **Vendor renewal watch** | Finds contracts approaching renewal with the owner, cost, and notice date | Creates time to cut or renegotiate spend | The vendor decision | | **Receivables follow-up** | Finds past-due invoices with no payment or dispute recorded | Protects cash already earned | The customer relationship and final message | | **Reporting input check** | Finds missing inputs and assembles submitted numbers | Removes the weekly chase and copy-paste | What the numbers mean | | **Close readiness check** | Reads the checklist and names open items, evidence, and owners | Shortens the exception list | Accounting judgment and sign-off | Each Runner works independently. Runners can read the same source apps, but they don't pass work or results to one another automatically. The safest first AI employee doesn't move money. They tell you where to look. ## Start with: invoice check ```text theme={null} Create a Runner named Invoice Check. Every Friday at 9:00am in my timezone, review vendor invoices received in the last 7 days in [Gmail or Drive folder], active agreements in [Drive folder], and the invoice tracking sheet in [Google Sheet]. For each invoice that isn't already marked reviewed in the tracking sheet, extract the vendor, invoice number, invoice date, service period, line items, total, currency, and payment terms shown in the source. Compare those fields with the matching active agreement and tracking sheet. Flag only: - a possible duplicate invoice number or amount - a rate, quantity, or total that doesn't match the agreement - a charge outside the agreement dates or scope - missing agreement, purchase reference, or required field - payment terms that changed Cite the invoice and agreement used for every flag. If a document is missing or unreadable, say so. Don't approve, reject, pay, or edit an accounting record. Return a table for review and a short list of questions for the owner. Before you create the Runner, show me the final name, prompt, schedule, and timezone. Wait for my confirmation. ``` **A user-created Runner becomes active as soon as it's created.** Review its schedule and permissions, then click **Run now**. Run it against a small set first. Five invoices you already checked will tell you more than a perfect-looking prompt. ## Add the rules to the saved prompt Put these in the Runner's saved prompt: * Where current agreements and approved vendor records live. * Required fields and matching tolerances. * Escalation timing. * Actions Runner must never take without a person. Use Project context for background such as vendor owners and business-unit names. Keep invoice, contract, and checklist status in the source files. Each run starts in a separate session and doesn't automatically receive the last run's output. Start every financial write on **Ask**. In many cases, the right permission is **Off**. Runner can collect and compare without approving anything. ## Other jobs to schedule ### Vendor renewal watch Run Monday morning. Find agreements renewing in the next 60 days, then show owner, current cost, notice deadline, and any renewal email. Draft the question. Don't renew or cancel. ### Receivables follow-up Run twice a week. Read the approved open-invoice list and matching customer thread. Draft a follow-up only when the source says the invoice is past due and no payment or dispute is recorded. ### Weekly operating report Run after the reporting deadline. Pull submitted numbers into the agreed format, identify missing inputs, and draft reminders for the owners. Don't invent a number to complete the page. ### Close readiness Run each weekday morning. If the close checklist isn't active, stop. During close, list open items and missing support, then name the source or owner shown. Keep reconciliations and sign-off with finance. ## The useful result is a shorter exception list Runner doesn't need to "run finance." It needs to take a long list and show only the items that deserve a human decision. Measure exceptions caught before approval and the report assembly you no longer do by hand. *** ## What's next? Find the next job tied to cost, time, or risk. Link the sources where invoices, agreements, and trackers live. Keep high-stakes writes behind a person. # Runner for owners and executives Source: https://guides.runner.now/use-cases/founders Add AI employees around revenue, customer risk, and operating decisions. A morning brief is useful. A brief that tells you which proposal went quiet, which customer promise is slipping, and which number needs a decision is much more useful. That's the move with Runner. Start with a Personal Assistant you already understand. Then add AI employees around the parts of the business that make or lose money. Writes follow your [tool permissions](/connections/tool-permissions). **Ask** stages an **Action Card** for review. **Allow** acts without asking. **Off** blocks the tool unless a user-created Runner has its own override. ## Recurring jobs for owners and executives | AI employee | What they own | Why the job matters | You stay focused on | | -------------------------- | ---------------------------------------------------------------------------- | ------------------------------------------------ | --------------------------------- | | **Executive brief** | Pulls the few revenue, customer, cost, and delivery items that need you | Gives your attention back before the day starts | Priority and judgment | | **Revenue follow-through** | Finds proposals and deals with no reply or next step | Keeps revenue already in motion from going quiet | The conversation and the close | | **Customer risk** | Finds unresolved questions, missed commitments, and renewals with no plan | Surfaces risk while there is still time to act | The relationship and the decision | | **Operating review** | Assembles current metrics, open projects, and blockers | Turns report building into decision time | Strategy and tradeoffs | | **Board update** | Builds a first draft from approved metrics, milestones, and the prior format | Removes the blank page and missing-number chase | The story and final message | Each Runner works independently. Runners can read the same source apps, but they don't pass work or results to one another automatically. The point isn't a prettier summary. It's catching the thing you would have found next Thursday and seeing it today. ## Start with: executive brief ```text theme={null} Create a Runner named Executive Brief. Every weekday at 7:30am in my timezone, review my connected Gmail, Google Calendar, [CRM], [project tracker], and the current operating metrics in [Google Sheet]. Only include items that need my attention in the next 7 days: 1. Revenue: new leads, proposals, or open deals with no clear next step. 2. Customers: commitments, risks, or questions waiting on us. 3. Cost: unusual changes or deadlines shown in the sources. 4. Delivery: blocked work, missed dates, or decisions waiting on me. 5. Today: meetings that need preparation. For every item, cite the source, explain why it matters, name the owner if one is shown, and give me one recommended next step. Don't invent a cause, owner, deadline, or dollar value. If the source is unclear, say so. Don't send messages or update records. Draft anything useful for my review. Return no more than 10 items. Rank them by money at risk and how soon I need to act. Before you create it, show me the final name, prompt, schedule, and timezone. Wait for my confirmation. ``` **A user-created Runner becomes active as soon as it's created.** Review its schedule and permissions, then run it once. If the list is full of noise, don't ask Runner to "be smarter." Tighten the sources, thresholds, and exclusions in the saved prompt. ## Add the business context Create a Project for shared background: * Current priorities and the metrics you actually care about. * Who owns which part of the business. * Where the CRM, project tracker, and reporting formats live. Put the definitions of an at-risk customer or stalled deal in the saved prompt. Put decisions Runner must never make there too, then enforce hard action boundaries with **Ask** or **Off**. Your CRM, inbox, project tracker, and metrics sheet hold live status. Each run starts in a separate session, so don't rely on yesterday's output being there tomorrow. Keep write tools on **Ask** while the Runner is learning your business. A suggested email can wait in an Action Card. The final decision is still yours. ## Other jobs to schedule ### Proposal and pipeline follow-through Check open revenue every weekday morning. Find proposals with no reply and deals with no next step. Stage a specific follow-up using the real thread and account context. ### Weekly operating review Run Friday afternoon. Compare the approved metrics with the prior period, pull blockers from the project tracker, and draft the questions that need an answer. Don't claim a cause unless a source states it. ### Customer risk review Run twice a week. Look for unresolved customer questions, missed commitments, open high-priority issues, and renewals approaching. Rank the list, show the owner recorded in the source, and flag anything with no owner for a person to assign. ### Board or investor update assembly Run on the date you normally start each monthly or quarterly update. Pull approved metrics and milestones into the existing format. Flag every missing number instead of guessing. ## Build toward the business Your Personal Assistant can give you a better morning. These Runners surface revenue, customer, and operating decisions before they disappear between meetings. Measure what reached you in time to act. *** ## What's next? Follow proposals and client delivery from source to next step. Find the next recurring job with a number attached. Put one valuable job on a schedule. # Runner for marketing and growth teams Source: https://guides.runner.now/use-cases/marketing-and-growth Add spend watch, pipeline review, lead follow-up, and content delivery. A campaign report can be perfectly accurate and still be useless. Clicks went up. CPA went down. Great. Did any of it create pipeline? Is one campaign burning through the week's budget? Is a high-value lead still waiting for a response? Give that job to an AI employee and put it on a schedule. Writes follow your [tool permissions](/connections/tool-permissions). Keep budget changes, CRM updates, and posts on **Ask** while you test. Runner stages each write as an **Action Card**. **Allow** acts without asking. **Off** blocks the tool unless the Runner has its own override. ## Recurring jobs for growth | AI employee | What they own | Why the job matters | You stay focused on | | ------------------------------- | ------------------------------------------------------------------------------ | ---------------------------------------- | --------------------------------- | | **Spend watch** | Checks campaign results against approved targets and CRM outcomes | Catches waste before month-end | The budget decision | | **Campaign-to-pipeline review** | Matches campaign and CRM records and calls out gaps | Moves the conversation past clicks | Attribution judgment and strategy | | **Lead aging** | Finds marketing leads with no owner or first response | Protects demand you already paid for | Qualification and follow-through | | **Content delivery** | Finds blocked, overdue, or ownerless work | Keeps production moving | Taste and editorial judgment | | **Weekly growth review** | Assembles current metrics, experiments, spend exceptions, and pipeline changes | Starts the meeting with current evidence | What to try next | Each Runner works independently. Runners can read the same source apps, but they don't pass work or results to one another automatically. Runner can compare the sources and apply your rules. It shouldn't invent causality. A metric moving after a campaign launched doesn't prove the campaign caused it. ## Start with: spend watch ```text theme={null} Create a Runner named Spend Watch. Every weekday at 8:00am in my timezone, review campaign performance in [connected ad source or Google Sheet], our approved targets in [Google Sheet], and pipeline outcomes in [HubSpot or Salesforce]. Only include a campaign when one of these is true: - daily spend is more than 20% above its approved pace - cost per qualified lead is more than 20% above target - spend increased but the connected CRM shows no matching qualified leads - required campaign or source data is missing For each exception, show campaign, spend, target, variance, qualified leads or pipeline shown in the CRM, and the exact source period. Separate facts from possible explanations. Don't claim why a metric changed unless a source says it. Recommend one thing for a person to inspect. Don't change budgets, pause ads, post messages, or update the CRM. Draft a short team summary for my review. If nothing is outside the rules, say that clearly and stop. Before you create the Runner, show me the final name, prompt, schedule, and timezone. Wait for my confirmation. ``` **A user-created Runner becomes active as soon as it's created.** Review its schedule and permissions, then click **Run now**. Twenty percent is an example. Use a threshold your team already trusts. ## Add the definitions Use a marketing Project for shared background: * Where the current numbers live. * What your team means by a qualified lead or pipeline opportunity. * Naming conventions for channels and campaigns. Put approved targets, thresholds, reporting windows, timezone, and prohibited changes in the saved prompt. Use **Ask** or **Off** for changes that must wait for a person. The ad source, CRM, and reporting sheet hold the live numbers. Previous run output doesn't carry forward automatically. Run the first few checks with writes on **Ask**. If the same false positive appears twice, fix the rule in the saved prompt. Don't just correct one session and hope the next run knows. ## Other jobs to schedule ### Campaign-to-pipeline review Run Monday morning. Match campaign and lead records using the identifiers your data actually provides. Show matched pipeline, unmatched records, and missing fields. Don't turn correlation into attribution. ### Lead aging check Run each weekday morning. Find new marketing leads with no owner or first response, then stage an alert or CRM update. Stop when the source record shows the lead is handled. ### Content bottleneck check Run each weekday morning. Find content due in the next seven days with no owner, blocked status, or overdue dependency. Draft the nudge, then let a person decide what moves. ### Weekly growth review Run Friday afternoon. Pull the approved metrics, current experiments, spend exceptions, and pipeline changes into one short report. Flag missing data instead of smoothing over it. ## A report isn't the outcome The goal is to catch wasted spend early, follow up on demand, and start the team meeting with current evidence instead of rebuilding the report. *** ## What's next? Follow demand from first response to next step. Find the next job tied to spend, pipeline, or time. Keep budget and CRM changes behind review. # Build your fleet Source: https://guides.runner.now/use-cases/overview Give your highest-value jobs to AI employees who work on a schedule across your apps. Runner lets you build a fleet of AI employees for your highest-value jobs. One follows every proposal. One watches spend. One catches a client commitment before it slips. One gets your day in order before the first meeting. Each Runner is one AI employee with a saved job, a schedule, named sources, and clear rules for when to bring work back to you. They handle the repetition. You keep the judgment, relationships, taste, and decisions that only you can make. Writes follow your [tool permissions](/connections/tool-permissions). **Ask** stages an **Action Card** for review. **Allow** acts without asking. **Off** blocks the tool unless a user-created Runner has its own override. ## Every job needs to earn its place Don't add another Runner because the demo looks clever. Add one because its job does at least one of these: * Drives or protects revenue. * Removes a real cost. * Saves enough time to matter. * Keeps a customer promise from slipping. The job should repeat. The sources should exist. You should be able to tell when it's done. And you should know what stays in your hands. ## A fleet built around real work | AI employee | What they own | Why the job matters | You stay focused on | | ---------------------- | ----------------------------------------------------------------------------------------- | -------------------------------------------- | ---------------------------------- | | **Proposal follow-up** | Checks the CRM and email for proposals waiting on a reply, then prepares the next message | Protects revenue already in motion | The relationship and the close | | **Lead response** | Finds untouched inquiries, reads the account context, and prepares the first response | Shortens the gap between interest and action | Qualification and the conversation | | **Cost watch** | Checks invoices, renewals, and spend against the rules you set | Finds waste and exceptions before month-end | The financial decision | | **Client delivery** | Finds blocked work, missing approvals, and promises without an owner | Protects margin and client trust | Tradeoffs and client relationships | | **Executive brief** | Pulls the few revenue, cost, customer, and delivery decisions that need you | Gives you time and attention back | Leadership and judgment | That's the lens for every use case in this section. You're not collecting things Runner *can* do. You're deciding which high-value jobs belong in the fleet. Each Runner works independently. Runners can use the same Project context, but they don't automatically pass work or results to one another. You decide how their jobs fit together. ## Start with your own day Your first AI employee can be a Personal Assistant. They read Gmail and Calendar, then give you a useful plan every morning. The job is familiar, the result is easy to check, and you learn how schedules and permissions work. Then move into the business. Add the AI employee who can protect the most revenue, remove the clearest cost, or save the most time. Personal: *What needs my attention today?* Business: *Which proposals have waited three business days without a reply, and what should we send next?* The setup is the same. The business job has a clearer number attached. ## Build the fleet your work needs Proposal follow-up, lead response, client onboarding, and delivery support. Executive briefings, revenue follow-through, customer risk, and operating reviews. Lead response, proposal follow-up, pipeline inspection, and renewals. Lead response, showing prep, seller updates, and pipeline follow-through. Spend watch, campaign-to-pipeline review, lead aging, and content delivery. Invoice checks, vendor renewals, receivables follow-up, and reporting. Customer impact checks, delivery risk, bug intake, and release summaries. A Personal Assistant, family planning, renewals, and open-loop follow-up. ## Keep the human work Runner should handle repeatable checks and bring you anything that needs judgment or approval. Keep these with a person: * The relationship. * The final high-stakes decision. * Taste, strategy, and tradeoffs. * Anything legal, financial, or hard to undo. *** ## What's next? Find the job with the clearest revenue, cost, or time attached. Give one AI employee a job, schedule, sources, rules, and tool permissions. # Runner for your personal life Source: https://guides.runner.now/use-cases/personal-and-family Build a fleet of AI employees for your calendar, commitments, renewals, and family plans. Your personal life already has workflows. They're just living in your head. Check the school email. Remember the dentist. Notice the subscription renewal. Look at the calendar before promising Saturday afternoon. None of it is hard. It just keeps coming back. Personal work is a good place to add your first AI employee because you already know what right looks like. Writes follow your [tool permissions](/connections/tool-permissions). Keep messages, calendar changes, purchases, cancellations, and bookings on **Ask** while you test. Browser checkouts and submissions add a final confirmation before they're submitted. ## Recurring jobs for personal life | AI employee | What they own | Why the job matters | You stay focused on | | ---------------------- | ------------------------------------------------------------------------- | ----------------------------------------- | ----------------------------- | | **Personal Assistant** | Reads Gmail and Calendar and gives you a short plan for the day | Saves morning planning time | Your priorities and choices | | **Follow-up check** | Finds messages and commitments still waiting on you | Keeps small promises from becoming stress | The relationship and reply | | **Family planner** | Pulls the next week of events, appointments, and missing details together | Cuts the Sunday-night scramble | Time with the people involved | | **Renewal watch** | Finds receipts and renewal notices with the amount and date shown | Creates a chance to stop waste | The keep-or-cancel decision | | **Trip researcher** | Compares options against the dates and preferences you give it | Cuts down the tab switching | The experience you want | Each Runner works independently. Runners can read the same source apps, but they don't pass work or results to one another automatically. Start with the Personal Assistant. This first AI employee gives you a useful result every morning while you learn how connected apps, schedules, and permissions work. ## Start with: Personal Assistant ```text theme={null} Create a Personal Assistant Runner for me. Every day at 9:00am in my timezone, review my connected Gmail and Google Calendar. Give me a short plan with: 1. Three items that may deserve my attention, with the source and why. 2. Meetings that need preparation. 3. Commitments and open loops that need attention. 4. Schedule conflicts, deadlines, or important messages I may have missed. Cite the email or event behind each item. If the source is unclear, say so. Don't send messages, change my calendar, buy, book, or cancel anything. Draft or suggest the next step for my review. Before you create it, show me the final name, prompt, schedule, and timezone. Wait for my confirmation. ``` **A user-created Runner becomes active as soon as it's created.** Review its schedule and permissions, then click **Run now**. The full walkthrough is in [Quickstart](/quickstart), including Gmail and Calendar permissions, testing with **Run now**, and saving feedback where it lasts. ## Save what should stick Put stable preferences in [Personal Memory](/using-runner/memory): * How early you want to hear about an event. * Travel, food, and transportation preferences. * Names and relationships you want Runner to understand. * The tone you use for personal messages. Put standing instructions in the saved prompt. Use a Personal Project for background such as names and source locations. Keep live dates, replies, and charges in Calendar, Gmail, or the tracking sheet Runner reads. Each run starts in a separate session. Last week's output doesn't automatically come along. If a fact should last, save it to Personal Memory. If a rule should apply every time, put it in the Runner's prompt. ## Other jobs to schedule ### Open-loop check Run at 4:30pm each weekday. Look for messages you sent that still need a reply and commitments due soon. Draft the next follow-up, but don't send it. ### Sunday family preview Run Sunday at 6pm. Read the next seven days of Calendar and relevant family email. List appointments, school events, travel, conflicts, and anything missing a time or place. ### Renewal review Run on the first of each month. Find renewal notices and recurring receipts in email, then show the amount and renewal date stated in the source. A receipt doesn't prove whether you still use something, so ask before recommending a cancellation. ### Trip research Run when you're planning, or on a schedule while prices matter. Compare options against the dates and preferences you give it. Runner shows the final browser purchase or booking action and asks before submitting. ## Reuse the pattern at work A morning plan gives you time back. A renewal review can save a little money. Once those work, take the same structure into your business: clear sources, clear rules, clear review points, and a schedule. That's how one useful AI employee becomes the start of a small fleet. *** ## What's next? Connect Gmail and Calendar, then run it once. Find the next personal or business job worth scheduling. Save facts and preferences that should carry into future sessions. # Runner for product and engineering teams Source: https://guides.runner.now/use-cases/product-and-engineering Add customer-impact checks, delivery risk, bug intake, and release summaries. A metric drops on Tuesday. The team notices Friday. By then you've spent three days sending more users into the same problem. Opening Mixpanel or GitHub wasn't the expensive part. Nobody was checking the metric against open issues and recent releases. That's a good job for an AI employee. Writes follow your [tool permissions](/connections/tool-permissions). Keep issue creation, status changes, and team posts on **Ask** while you test. Runner stages each write as an **Action Card**. **Allow** acts without asking. **Off** blocks the tool unless the Runner has its own override. ## Recurring jobs for customer impact | AI employee | What they own | Why the job matters | You stay focused on | | ------------------------- | ------------------------------------------------------------------ | ------------------------------------------ | -------------------------------- | | **Customer-impact check** | Checks approved metrics, open issues, and recent releases | Catches customer problems earlier | Root-cause judgment and priority | | **Delivery risk** | Finds blocked work, missing owners, and overdue dependencies | Keeps important work from slipping quietly | Tradeoffs and team leadership | | **Bug intake** | Groups new reports and checks for existing issues | Cuts duplicate triage work | Severity and priority | | **Release summary** | Builds a draft from merged pull requests and completed issues | Removes the release-note chase | The customer story | | **Product review** | Connects shipped work with approved usage metrics and open reports | Brings evidence to the roadmap discussion | Product judgment and the bet | Each Runner works independently. Runners can read the same source apps, but they don't pass work or results to one another automatically. Runner can connect the evidence. It can't prove that one deploy caused a metric change just because the timestamps are close. ## Start with: customer-impact check ```text theme={null} Create a Runner named Customer Impact Check. Every weekday at 8:30am in my timezone, review the approved product metrics in Mixpanel, high-priority open issues in Linear, and recent releases in GitHub. Use these thresholds in every run: - Key metric ranges: [metric and approved range] - Funnel change: [percentage] over [comparison window] - High-priority issue definition: [definition] Only include: - a key metric outside the saved range - a funnel step that changed more than the saved threshold - a new high-priority issue with no owner or next step - a release with a related open issue or documented customer report For each item, cite the metric period, event or funnel, Linear issue, and GitHub release used. Separate facts from possible explanations. A nearby release is a lead to investigate, not proof of cause. Recommend one next check. Don't roll back code, change issue status, assign an owner, or post to Slack. Draft a Linear issue or team update only when the sources contain enough detail for a person to review it. If nothing matches, say that clearly and stop. Before you create the Runner, show me the final name, prompt, schedule, and timezone. Wait for my confirmation. ``` **A user-created Runner becomes active as soon as it's created.** Open it, assign the Project that contains the metric definitions and source locations, confirm the saved thresholds, and review its schedule and permissions, then click **Run now**. Use a schedule that matches how quickly your team can respond. A daily check nobody reads is worse than a twice-weekly check somebody owns. ## Keep the rules in the saved prompt Use a product Project for shared background: * The names and source locations of the metrics that matter. * Severity vocabulary and team ownership. * Release and issue naming conventions. Put thresholds, comparison windows, timezone, exclusions, and prohibited actions in the saved prompt. Metrics, issues, and releases hold live state. Previous run output doesn't carry forward automatically. Review the first few results with write tools on **Ask**. Tighten a noisy threshold in the saved prompt. Remove a vanity metric. Correct inaccurate background in the Project. Don't leave either fix inside one run. ## Other jobs to schedule ### Delivery risk Run each weekday morning. Find projects due in the next 14 days with blocked work, missing owners, overdue dependencies, or no recent update. Name the evidence and draft the question for the owner. ### Bug intake Run each weekday morning. Group new reports by the product area and symptoms shown in the source. Search existing Linear issues before drafting a new one. A person still chooses severity and priority. ### Release summary Run Friday after your normal release window. If no release appears in the source from the past seven days, stop. Otherwise, pull merged work and completed issues into a customer-facing and internal draft. Don't claim impact before the data exists. ### Weekly product review Run Friday afternoon. Compare shipped work with the approved metrics and open customer reports. Call out missing or inconclusive data instead of forcing a win. ## Protect the work that already shipped The goal isn't more tickets. It's catching a customer problem earlier, keeping an important project from slipping, and giving the team evidence before it makes a decision. That saves engineering time and protects revenue without pretending the system knows more than the sources do. *** ## What's next? Link Linear, GitHub, Mixpanel, and the sources your team trusts. Find the next job tied to customer or business impact. Put one scheduled check behind clear rules. # Runner for professional services Source: https://guides.runner.now/use-cases/professional-services Build a fleet of AI employees for proposal follow-up, client onboarding, and delivery. A \$25,000 proposal goes out Tuesday. Friday arrives. Everyone assumes someone followed up. Nobody did. That's the kind of job an AI employee should own. Not writing another generic email on command. Checking the real pipeline every morning, finding the proposal that went quiet, reading the thread, and preparing the right next step. For agencies, consultancies, studios, and development shops, the first AI employees should sit close to revenue or client delivery. Writes follow your [tool permissions](/connections/tool-permissions). Keep email sending on **Ask** while you test. Runner stages each send as an **Action Card** for review. **Allow** sends without asking. **Off** blocks the tool unless the Runner has its own override. ## Recurring jobs for professional services | AI employee | What they own | Why the job matters | You stay focused on | | ---------------------- | ------------------------------------------------------------------------------------- | ------------------------------------------------- | ---------------------------------------- | | **Proposal follow-up** | Checks CRM and email for proposals waiting on a reply, then prepares the next message | Keeps good work from going quiet | The relationship and the close | | **New lead response** | Finds untouched inquiries, reads the account context, and prepares the first response | Shortens the gap between interest and action | Qualification and the conversation | | **Client onboarding** | Turns a signed deal into a welcome draft, checklist, and missing-input list | Gets delivery moving without another manual chase | Expectations and the client relationship | | **Delivery risk** | Finds blocked work, missing approvals, and promises without an owner | Protects margin and client trust | Tradeoffs and hard conversations | | **Account review** | Surfaces wins, risks, renewals, and the next useful conversation | Protects renewals and follow-on work | The relationship and judgment | Each Runner works independently. Runners can read the same source apps, but they don't pass work or results to one another automatically. You don't need all five AI employees. Add the one with the clearest number attached. ## Start with: proposal follow-up This works because the state already lives in Gmail and your CRM. Every run can see who replied, what was promised, and whether the opportunity is still open. ```text theme={null} Create a Runner named Proposal Follow-up. Every weekday at 8:00am in my timezone, review open proposals in [HubSpot or Salesforce] and the matching Gmail threads. Include proposals that: - were sent 3 to 14 business days ago - have no client reply after the proposal was sent - are not marked won, lost, paused, or do not contact - no unresolved follow-up Action Card is already present for the proposal Read the proposal in Google Drive and the full email thread before writing. For each proposal, prepare a short follow-up with one specific next step. Don't invent urgency, discounts, or facts that aren't in the source material. Don't follow up with the same person more than once every 5 business days. Stage each send as an Action Card. Then return a short report with the client, proposal value shown in the CRM or "not recorded," days waiting, last contact, and proposed next step. If nobody needs a follow-up, say that clearly and stop. Before you create it, show me the final name, prompt, schedule, and timezone. Wait for my confirmation. ``` **A user-created Runner becomes active as soon as it's created.** Open it, review the saved prompt and schedule, then click **Run now** for the first test. Turn on **Carry unresolved Action Cards into new runs** so a still-pending follow-up can appear in the next run. A pending card still expires 24 hours after it was staged. ## Save the job rules Use a Project for background such as your offer, audience, account owners, and source locations. Put these in the Runner's saved prompt: * The voice to use with prospects. * Deal stages and exclusion rules. * The minimum gap between follow-ups. * What needs a partner or account owner to review. Keep the live state in Gmail and the CRM. That's what tells tomorrow's run whether the client replied. The previous run's output doesn't carry forward automatically. Start sends on **Ask**. Read the first ten Action Cards. Fix the saved prompt when you see the same mistake twice. Move a send tool to **Allow** only when you're comfortable letting that exact action happen without review. ## Other jobs to schedule ### New lead response Run at the start and end of each workday. Find new inquiries with no owner or first reply, read the account context from connected sources, and stage a response. Stop when the CRM or email thread shows someone already handled it. ### Client onboarding Run each weekday morning. Find newly won work, then prepare the welcome email, kickoff checklist, and list of missing inputs. Keep contract, payment, and scope decisions with a person. ### Delivery risk Run before the team starts work. Find deliverables due in the next seven days, blocked tasks, and client questions waiting more than two business days. Draft the nudge and name the owner. ### Weekly account review Run Friday afternoon. Pull what shipped, what slipped, open client questions, and the next milestone into a clean account summary. No green status based on vibes. Use the work and messages that actually exist. ## The point isn't more activity A proposal follow-up Runner isn't valuable because it writes five emails. Measure how often a proposal waits ten days without a next step. Fix that gap, then build the next Runner. *** ## What's next? Save the job, schedule, sources, and review rules. Go deeper on lead response, pipeline gaps, and next steps. Decide which actions wait for review. # Runner for real estate agents Source: https://guides.runner.now/use-cases/real-estate-agents Add lead response, showing prep, seller updates, and pipeline follow-through. A lead arrives at 7:12pm. You mean to answer after dinner. By 9am they're already talking to somebody else. The valuable work isn't "write an email." It's checking for the lead, reading the inquiry, checking your calendar, preparing the right response, and stopping once a person has replied. That's one AI employee with a clear job. Writes follow your [tool permissions](/connections/tool-permissions). Keep messages, CRM updates, bookings, and submissions on **Ask** while you test. Runner stages each write as an **Action Card**. Browser bookings and submissions add a final confirmation before they're submitted. ## Recurring jobs around a transaction | AI employee | What they own | Why the job matters | You stay focused on | | ------------------- | ------------------------------------------------------------------------------- | ------------------------------------------------------- | ---------------------------------- | | **Lead response** | Finds new buyer and seller inquiries and prepares a reply from the real context | Shortens the gap before another agent responds | The relationship and qualification | | **Lead follow-up** | Finds open leads with no reply or next step | Keeps serious conversations from going quiet | The personal follow-up | | **Showing prep** | Pulls the client thread, calendar details, and property material into one brief | Cuts prep before every showing | The client experience | | **Seller update** | Drafts an update from confirmed activity, feedback, and open questions | Keeps communication moving without rebuilding the story | The relationship and hard message | | **Pipeline review** | Finds clients and leads missing an owner, reply, or next step | Shows where a transaction may stall | Priority and outreach | Each Runner works independently. Runners can read the same source apps, but they don't pass work or results to one another automatically. Don't give one AI employee the whole transaction. Start with the lead that needs a reply. ## Start with: lead response ```text theme={null} Create a Runner named New Lead Response. Every weekday at 8:00am and 4:00pm in my timezone, review new buyer and seller inquiries in my connected Gmail and [CRM]. Include an inquiry only when: - it arrived after 8am that day on the 4pm run, or after the most recent weekday at 4pm on the 8am run - no agent is shown as the owner, or I am the owner - nobody has replied after the inquiry - the person is not marked do not contact - no unresolved reply Action Card is already present for the inquiry Read the full inquiry and matching CRM record. Draft a short response that answers only what the sources support. If they asked to talk or tour, read my Google Calendar and suggest two available windows. Don't promise availability, pricing, property facts, representation, or a booking that isn't confirmed. Stage the reply as an Action Card. Return the lead, inquiry time, source, assigned owner, and proposed next step. If there are no untouched inquiries, say that clearly and stop. Never use or infer protected characteristics when ranking, routing, or writing to a lead. Before you create the Runner, show me the final name, prompt, schedule, and timezone. Wait for my confirmation. ``` **A user-created Runner becomes active as soon as it's created.** Review its schedule and permissions, then click **Run now**. A Runner that fires twice a day uses more of your plan. Start once a day if a second check isn't worth the cost. Turn on **Carry unresolved Action Cards into new runs** so a pending reply can appear at the next check instead of being drafted twice. The card still expires 24 hours after it was staged. Run it with a few test inquiries first. Read every reply. Fast and wrong isn't better than slow. ## Save the job rules Use a real estate Project for background such as service area, property types, lead owners, and source locations. Put these in the Runner's saved prompt: * Working hours and routing rules. * Approved voice and response templates. * The minimum gap between follow-ups. * Fair housing, brokerage, and legal boundaries. * Actions that always need you or your broker. Keep live lead status in the CRM and conversation status in email. Each run reads those sources again. It doesn't automatically remember the prior run's output. Start every send and CRM write on **Ask**. Keep offer, contract, booking, and submission decisions behind a person. ## Other jobs to schedule ### Lead follow-up Run every weekday morning. Find open leads with no reply or next step after the gap you set. Read the actual thread before drafting anything. Stop when the source says the person replied, opted out, or the lead closed. ### Showing prep Run each morning. For today's showings, gather the client thread, calendar details, and property documents you provided. List missing instructions instead of filling the gaps with guesses. ### Seller update Run every weekday at 5pm. If Calendar shows no showing that day, stop. Otherwise, draft a short update from confirmed activity, feedback, and open questions. Don't invent buyer interest or market signals. ### Weekly pipeline review Run Friday afternoon. Pull current CRM status, recent client threads, and next week's dates into one list. Flag missing owners and next steps. Don't rank people by vague labels such as "hot" unless your CRM contains an explicit rule. ## Speed matters. So does trust. The first win is simple: new leads get checked on schedule, the reply uses real context, and nothing sends before you're ready. Then you add the next Runner. *** ## What's next? Link Gmail, Calendar, and the CRM you use. Find the next job closest to revenue or cost. Keep messages and submissions behind review. # Runner for sales and RevOps teams Source: https://guides.runner.now/use-cases/sales-and-revops Build a fleet of AI employees for leads, proposals, pipeline gaps, and renewals. A full pipeline can still be a dead pipeline. The problem usually isn't the number of records. It's the gaps: a new lead with no reply, a proposal with no follow-up, a deal with no next step, or a renewal nobody owns. Those gaps repeat every week. They are good jobs for AI employees. Writes follow your [tool permissions](/connections/tool-permissions). Keep CRM updates and email sending on **Ask** while you test. Runner stages each write as an **Action Card**. **Allow** acts without asking. **Off** blocks the tool unless the Runner has its own override. ## Recurring jobs for revenue | AI employee | What they own | Why the job matters | You stay focused on | | ----------------------- | ---------------------------------------------------------------------------- | -------------------------------------------- | ---------------------------------- | | **Lead response** | Finds new inquiries with no owner or reply, then prepares the first response | Protects the demand you already earned | Qualification and the conversation | | **Proposal follow-up** | Checks sent proposals and matching threads, then prepares the next message | Keeps serious opportunities from going quiet | Negotiation and the close | | **Pipeline inspection** | Finds deals with no next step, stale activity, or overdue tasks | Shows where revenue is stuck | Forecast judgment and coaching | | **Renewal watch** | Finds approaching renewals with open issues or no plan | Protects existing revenue | The customer relationship | | **Sales reporting** | Pulls won, lost, slipped, and missing CRM data into one report | Removes manual report assembly | The call on what changes next | Each Runner works independently. Runners can read the same source apps, but they don't pass work or results to one another automatically. Don't build one AI employee to "do sales." Give them one gap you can see and measure. The 8am and 4pm lead check uses more of your plan. Start with the morning check until the second one is worth the cost. ## Start with: pipeline inspection ```text theme={null} Create a Runner named Pipeline Inspector. Every Monday and Thursday at 8:00am in my timezone, review open opportunities in [HubSpot or Salesforce] and the matching Gmail threads. Include a deal when at least one of these is true: - there is no next step or next-step date - the next-step date has passed - there has been no recorded activity for 7 days - a proposal was sent at least 3 business days ago with no client reply Exclude deals marked won, lost, paused, or do not contact. For each deal, read the CRM record and latest email thread. Return the deal, stage, value shown in the CRM, days since activity, owner, missing field, and one recommended next step. If a follow-up is appropriate, draft it from the real context and stage the send as an Action Card. Don't invent urgency, pricing, product claims, or a deadline. If no deals match, say that clearly and stop. Before you create the Runner, show me the final name, prompt, schedule, and timezone. Wait for my confirmation. ``` **A user-created Runner becomes active as soon as it's created.** Review its schedule and permissions, then click **Run now**. The result should be small enough to act on. Ten real gaps beat a 60-page pipeline export. ## Keep the sources reliable Use a sales Project for background such as the ideal customer, deal-stage names, ownership, product facts, and source locations. Put disqualifiers, stage requirements, follow-up timing, do-not-contact rules, approved claims, and pricing boundaries in the saved prompt. Keep deal status in the CRM and conversation status in email. The next run reads those sources again. It doesn't automatically receive the previous run's output. Start with writes on **Ask**. Review the first ten email and CRM Action Cards. If a rule keeps needing correction, change the saved prompt. Don't leave the fix in one run's session. ## Other jobs to schedule ### New lead response Run at the start and end of each workday. Find new inbound leads with no owner or reply, read the available company and CRM context, and stage a short first response. Stop as soon as the record or thread shows somebody handled it. ### Proposal follow-up Run each weekday morning. Check proposals sent 3 to 14 business days ago. Draft the next message only when the client hasn't replied and the opportunity is still open. ### Renewal watch Run twice a week. Find renewals in the next 60 days, then gather recent email, open issues, usage data that is actually connected, and the named owner. Surface missing plans early. ### Weekly sales report Run Friday afternoon. Pull what changed from the CRM, not from memory. Show won, lost, slipped, stale, and missing data. Keep forecast judgment with the sales owner. ## Measure the gap, not the output Don't count how many emails Runner drafted. Count leads that got a first response, proposals that received a next step, and open deals that stopped sitting without an owner. That's the revenue work. *** ## What's next? Build proposal follow-up and client onboarding Runners. Link your CRM and Gmail, then review their tools. Save the check, schedule, and review boundary. # Talking to Runner Source: https://guides.runner.now/using-runner/chatting Give Runner a clear brief, check the evidence, and turn useful work into a Workflow or Runner. A session is where you work with Runner. Brief the job, answer questions, check the evidence, make the call, and improve the result. Runner can find, compare, organize, and draft across the sources you give it. You keep the judgment, relationships, creativity, and leadership that need a person. Writes follow your [tool permissions](/connections/tool-permissions). **Ask** stages an **Action Card** for review. **Allow** acts without asking. **Off** blocks the tool unless a user-created Runner has its own override. ## Work together in a session | You ask | Runner handles | You decide | | ------------------------------------------------ | --------------------------------------------------------------------------------- | ------------------------------------------------------------------ | | *"Triage email from the last seven days"* | Reads, groups, cites, and drafts from the inbox | What matters, how a reply should land, and whether to send | | *"Prep me for tomorrow's customer meeting"* | Pulls the calendar, relevant threads, open commitments, and named Project sources | The goal, questions, relationship, and how to lead the room | | *"Check proposals waiting on a reply"* | Finds the gaps, reads the thread, and drafts the next step | Timing, price, negotiation, and the final message | | *"Compare these invoices with the signed terms"* | Shows mismatches and links the evidence | Whether it's correct, how to handle the vendor, and whether to pay | No special syntax. Say it like you'd say it to a capable colleague, then make the decisions that need you. *** ## Give the job the right sources } title="Gmail" /> } title="Calendar" /> } title="Slack" /> } title="Linear" /> } title="Notion" /> } title="Drive" /> Connect only what the job needs. Gmail and Calendar are enough for a Personal Assistant. Add Slack, Linear, or Notion when a different job needs them. [Connect your apps →](/get-started/connecting-accounts) *** ## Save what should carry forward Want a preference available in future sessions? Ask Runner to remember it: * *"When you draft emails for me, keep them short. Three sentences max."* * *"Always check my calendar before suggesting meeting times."* * *"Emails from Sarah and James are always high priority."* Then open **Memory** and make sure it was saved. Future sessions can use saved facts and preferences. The transcript itself doesn't become Memory. [See and edit Personal Memory →](/using-runner/memory) *** ## Voice and files **Voice input.** Click the mic icon to talk instead of type. Dictate a task from your phone on the way to the gate, close it, and check the result when you land. [Runner on your phone →](/get-started/runner-on-your-phone) **Attach files.** Drag a PDF, spreadsheet, or image into the conversation. *"Review this proposal and give me the three biggest concerns."* *** ## Sessions follow you Every session syncs across your devices. Start on your laptop, continue on your phone. Need context from another session? Ask Runner to read another active session you own and use it here. It can't send a message into or resume that other session. *** ## Give Runner a clear brief * **Give one clear outcome.** *"Check the email from Sarah about the Q2 budget"* beats *"check my email."* * **Name the sources.** Say which inbox, Project, folder, record, or site holds the truth. * **Set the review point.** Say what should stay a draft. Use **Ask** or **Off** for hard action boundaries. * **Correct the work.** Tell Runner what was wrong and what good looks like. Feedback in one session doesn't rewrite a scheduled Runner. For a lasting job change, update its saved prompt. Put durable personal facts and preferences in Personal Memory. ## When to use a session, Workflow, or Runner * Use a **session** to work through something once. * Save a **Workflow** when the procedure is useful again and you want to start it with `/slug`. * Create a **Runner** when one AI employee should own the job on a schedule. Test the work in a session first. Save the procedure as a Workflow when you want it again. Create a Runner when the result is useful, repeatable, and ready for a schedule. *** ## What's next? Save a useful procedure and invoke it with /slug. Put one valuable job on a schedule with rules and tool permissions. # Memory Source: https://guides.runner.now/using-runner/memory See, edit, and organize the facts and preferences Runner can use across sessions. Personal Memory is where Runner keeps durable facts and preferences your AI employees can use across jobs and sessions. * *"Keep my email drafts to three sentences max."* * *"Sarah and James are always high priority."* * *"I don't have a car. Transit options only."* Want one of those to carry forward? Ask Runner to remember it, then check **Memory** to make sure it was saved. A session transcript isn't automatically Memory. Neither is the output from a previous Runner run. The fact has to be written to Personal Memory before another session or run can use it there. ## Put each kind of context in the right place | What you want to keep | Where it belongs | | --------------------------------------------------- | ------------------------------------------------------- | | A personal fact or preference | **Personal Memory** | | A standing rule for one Runner | That Runner's saved **Prompt** | | Shared background, definitions, or source locations | **Project context** | | A reusable procedure you choose when to start | A **Workflow** | | Live deal, invoice, project, or customer status | The source app the next run can read | | One run's result | Its session, unless you save the durable part elsewhere | Memory isn't where every result belongs. Save only the durable part another session or run should use. ## See what Runner remembers Open **Memory** in the sidebar. Each memory is a document you can read, edit, or delete. Folders can be nested, so you can organize people, preferences, company context, or anything else that needs a home. If something is wrong, open it and fix it. You can also tell Runner what changed and ask it to update the memory. You can add one yourself too. Click **New memory**, give it a title and body, done. *** ## Memory keeps itself clean Memory that grows forever becomes a junk drawer. Runner does its own housekeeping: **It tidies.** Open **Memory** and click **Organize**. Runner files stray memories into folders in the background. It also runs a weekly cleanup to combine duplicates and keep the structure useful. **It stays fresh.** Runner periodically sweeps supported connected apps with read-only tools. Those sweeps can update Personal Memory, but they don't change the connected apps. *** ## Turn feedback into memory Correct the result, then say what should be remembered: > *"You recommended I drive there, but I don't have a car. Remember to use transit options next time."* Check **Memory** afterward. If the correction was saved, future sessions can use it without making you repeat yourself. *** ## What's next? Keep context that belongs to one Project with that Project. See what each Runner gets when a new run starts. # Choose an intelligence setting Source: https://guides.runner.now/using-runner/models Match the reasoning level to the job. Sessions use Auto, Normal, or Ultra. Runners use Normal or Ultra. Intelligence is the reasoning level you assign to the job. Most jobs should start on **Normal**. Move to **Ultra** only when the work needs deeper reasoning. ## Sessions have three settings | Setting | Use it for | | ---------- | -------------------------------------------------------------------------------------------- | | **Auto** | Let Runner choose for the task in front of it. | | **Normal** | Attention briefs, pipeline checks, summaries, scheduling, and everyday work. | | **Ultra** | Complex research, long documents, ambiguous comparisons, and work that needs more reasoning. | New sessions start on **Normal** unless you change **Default intelligence** under **Settings**, then **Behavior**. That preference applies to new sessions on the device where you changed it. You can still change the intelligence setting inside a session whenever the work changes. ## Scheduled Runners use Normal or Ultra A scheduled Runner can't use **Auto**. Choose **Normal** for routine work and **Ultra** when the job needs the extra depth. New Runners start on **Normal** unless you choose **Ultra** while setting one up. Changing intelligence doesn't change tool permissions. **Ultra** doesn't let Runner read, change, or send anything **Normal** couldn't. ## Choose what sits behind each setting Open **Settings**, then **Behavior** to choose the supported model behind **Normal** and **Ultra**. The available list can change as models are added or retired. The session picker stays simple. It shows **Auto**, **Normal**, and **Ultra** rather than a separate exact-model list. ## Keep an eye on usage **Ultra** uses more of your plan than **Normal**. That matters most for a Runner that runs often. Use **Normal** for recurring triage and summaries. Move to **Ultra** when the result needs it. Simple. *** ## What's next? Set **Normal** or **Ultra** for recurring work. Intelligence controls how Runner reasons. Permissions control what it can do. # Workflows Source: https://guides.runner.now/workflows/overview Save instructions once, then invoke them with /slug in any session or Runner. A Workflow saves instructions you want to reuse with `/slug`. It's a playbook, not an AI employee. Say you're always asking for the same inbox triage: archive newsletters, treat investors as VIPs, and keep drafts under three sentences. A Workflow captures that procedure once. Type `/inbox-triage` and Runner loads those instructions in that session. Writes follow your [tool permissions](/connections/tool-permissions). **Ask** stages an **Action Card** for review. **Allow** acts without asking. **Off** blocks the tool unless a user-created Runner has its own override. Coming from Runner Desktop? Workflows are what Skills became. Your synced Skills are already in your [Workflow Library](https://app.runner.now/app/workflows). [Migration details →](/legacy/runner-desktop) ## Create a Workflow Two ways, same as Runners: fill out the form, or have Runner build it with you. Find **Workflows** in your sidebar. Click **New workflow**, or choose **Create a workflow** to have Runner interview you and write it. Name it (the name becomes the slug: "Listing Email" becomes `/listing-email`) and write the step-by-step guidance Runner should follow. Be as opinionated as you are in real life. That's the point. Type `/listing-email` in any session, on any device. Add context right in the message: *"/listing-email for the house on Maple, emphasize the backyard."* Just did something in a session you'll want again? Say *"turn this into a workflow."* Runner extracts the procedure from the conversation and saves it. *** ## Workflows and Runners work together A Runner owns the job. A Workflow says how, the schedule says when, and the Project supplies background. A Runner assigned the Friday review can invoke `/weekly-review` at 4pm, using those instructions and the background from its Project. | You want | Use | | ---------------------------------------------------------- | ----------------------------------- | | "Do this the way I like, whenever I ask" | A Workflow | | "Do this every morning without me asking" | A Runner | | "Do this the way I like, every morning, without me asking" | A Runner that invokes your Workflow | *** ## Workflow ideas | Workflow | What's inside | | ---------------- | ------------------------------------------------------------------------- | | `/inbox-triage` | Your triage rules: what gets archived, who's VIP, how drafts should sound | | `/meeting-prep` | Your prep format: attendees, history, open commitments, talking points | | `/listing-email` | Your listing email structure, tone, and the details you always include | | `/review-pr` | Review a pull request the way you like, with your checklist | | `/weekly-report` | Your report template and where to pull the numbers from | *** ## What's next? Start with useful procedures, then make them yours. Put a recurring job and the right Workflow on a schedule. # Starter workflows Source: https://guides.runner.now/workflows/starter-workflows Workflows for jobs that protect revenue, cut costs, save time, or keep customer delivery moving. A Workflow is a playbook. It tells Runner how you want a job done. Try one in a session first. Save it as a Workflow when you want the same method again. Have a Runner invoke it when you want that job to run on a schedule. Writes follow your [tool permissions](/connections/tool-permissions). **Ask** stages an **Action Card** for review. **Allow** acts without asking. **Off** blocks the tool unless a user-created Runner has its own override. ## Find the right Workflow | The result you want | Start here | | --------------------------------------- | --------------------------------------------------- | | Protect a proposal or deal | [Proposal follow-up](#proposal-follow-up) | | Find holes in the pipeline | [Pipeline inspection](#pipeline-inspection) | | Catch a renewal before it slips | [Renewal watch](#renewal-watch) | | Stop an invoice mistake | [Invoice check](#invoice-check) | | Find recurring spend | [Spend review](#spend-review) | | Start the day knowing what matters | [Morning attention brief](#morning-attention-brief) | | Walk into a meeting prepared | [Meeting prep](#meeting-prep) | | Find delivery risk early | [Delivery risk check](#delivery-risk-check) | | Keep a client informed | [Weekly client update](#weekly-client-update) | | Clear the inbox without losing judgment | [Inbox triage](#inbox-triage) | ## Protect revenue ### Proposal follow-up **Job:** Find proposals that have waited more than three business days for a reply. **Value:** Protects work you've already paid to win. **Needs:** Gmail, plus the CRM or proposal tracker you use. **Output:** A ranked list with proposal value, last contact, days waiting, and a draft follow-up. **Human decision:** You choose the timing, message, pricing, and whether to send. **Use as:** A weekday Runner for an active sales pipeline, or a Workflow you invoke before pipeline review. > *"Find proposals sent more than three business days ago with no reply. Rank them by value. Read the thread, draft the next follow-up, and don't send anything."* ### Pipeline inspection **Job:** Find open deals with no next step, overdue follow-up, or no recent activity. **Value:** Protects revenue from quiet process gaps. **Needs:** Your CRM and, when useful, Gmail. **Output:** A short exception list with owner, stage, value, last activity, and missing next step. **Human decision:** The seller owns discovery, forecast judgment, pricing, negotiation, and the relationship. **Use as:** A weekday or Monday-morning Runner. > *"Check open deals for missing next steps, overdue follow-up, or no activity in seven days. Show only exceptions. Don't change records or contact anyone."* ### Renewal watch **Job:** Surface upcoming renewals and unresolved customer risks before the deadline. **Value:** Protects recurring revenue and customer trust. **Needs:** Your CRM, agreement folder, support source, or renewal tracker. **Output:** Renewals ranked by date and value, with visible risks and missing owners. **Human decision:** You own the customer conversation, commercial terms, and the renewal call. **Use as:** A weekly Runner. > *"Find renewals due in the next 60 days. Show contract value, owner, open customer risks, last contact, and the next decision needed. Don't infer health when the source is unclear."* ## Cut costs ### Invoice check **Job:** Compare incoming invoices with signed terms and flag mismatches. **Value:** Stops duplicate charges, wrong rates, and avoidable spend before payment. **Needs:** Gmail or the invoice folder, the signed-agreement folder, and your tracking sheet. **Output:** Exceptions with vendor, invoice, contract term, difference, and source links. **Human decision:** A person verifies the evidence, handles the vendor, and approves or pays. **Use as:** A daily or weekly Runner, depending on volume. > *"Check new invoices against the signed agreement and our tracker. Show only duplicates, rate mismatches, missing terms, or totals that don't reconcile. Never approve or pay anything."* ### Spend review **Job:** Find recurring charges, renewals, and spend that changed unexpectedly. **Value:** Removes waste you don't notice one receipt at a time. **Needs:** Gmail receipts, statements or a spend sheet, and any agreement folder you use. **Output:** A table with vendor, amount, frequency, last charge, change, and evidence. **Human decision:** You decide what the business still needs and whether to cancel or renegotiate. **Use as:** A monthly Runner or an on-demand Workflow. > *"Build a recurring-spend review from the sources I give you. Flag duplicates, price changes, and renewals due in the next 45 days. Don't cancel or contact a vendor."* ## Save time ### Morning attention brief **Job:** Turn Gmail and Google Calendar into a short plan for the day. **Value:** Saves the first part of your morning and makes missed commitments less likely. **Needs:** Gmail and Google Calendar. **Output:** Three candidate priorities with source evidence, meetings that need prep, open commitments, conflicts, and important messages. **Human decision:** You choose the priorities, make the calls, and reply to people. **Use as:** A 9am Personal Assistant Runner. The [Quickstart](/quickstart) builds this one. > *"Give me a short plan for today from Gmail and Google Calendar. Cite the source for each item. If the evidence is unclear, say so instead of guessing."* ### Meeting prep **Job:** Assemble the context you need before the next important meeting. **Value:** Saves research time while protecting the quality of the conversation. **Needs:** Google Calendar, Gmail, and any Project sources you name. **Output:** Attendees, recent history, open commitments, useful context, and questions to consider. **Human decision:** You choose the goals, read the room, build the relationship, and lead the meeting. **Use as:** A Workflow you invoke for selected meetings, or a scheduled Runner for a narrow calendar rule. > *"Prepare me for my next meeting. Use the calendar event, relevant email threads, and the Project context. Separate sourced facts from suggestions."* ### Inbox triage **Job:** Sort recent email into what needs you, what can wait, and what looks like noise. **Value:** Saves time while keeping sensitive replies under review. **Needs:** Gmail. **Output:** A short priority list, waiting-on list, and drafts for messages that need a reply. **Human decision:** You decide which relationships matter, how a reply should land, and what gets sent. **Use as:** An on-demand Workflow first. Schedule it only after the rules are stable. > *"Triage email from the last seven days. Show what needs me, what I'm waiting on, and what can wait. Draft important replies, but don't send or archive anything."* ## Protect customer delivery ### Delivery risk check **Job:** Find commitments with no owner, overdue work, blocked work, and client dependencies. **Value:** Protects delivery dates and the trust behind them. **Needs:** Your project tracker, Gmail, and the Project brief. **Output:** Exceptions ranked by customer impact, with owner, due date, blocker, and source. **Human decision:** The team owns priority, tradeoffs, promises, and the customer conversation. **Use as:** A weekday Runner for one delivery team or client Project. > *"Find delivery risks from the tracker and relevant email. Show only missing owners, overdue items, blockers, and commitments likely to slip. Cite the source. Don't change dates or contact the client."* ### Weekly client update **Job:** Draft a factual update from work that actually happened. **Value:** Saves reporting time and keeps client updates on schedule. **Needs:** Your project tracker, calendar, relevant email, and the client Project. **Output:** Progress, decisions, risks, next steps, owners, and dates, with gaps called out. **Human decision:** You choose the framing, handle sensitive news, and send the update. **Use as:** A Friday Runner with sending on **Ask**, or a Workflow you invoke after review. > *"Draft this week's client update from the sources in this Project. Separate completed work, decisions, risks, and next steps. Don't hide missing information and don't send it."* ## Put a useful Workflow on a schedule Run the procedure once in a session. Check the sources, edge cases, output, and which actions or decisions still need a person. Then choose: * Keep it in the **session** when the work happens once. * Save it as a **Workflow** when you want the same procedure on demand. * Have a **Runner** invoke it when the job should run on a schedule. A Workflow says how. The Runner owns the job. The schedule says when. The Project supplies background. Don't add another Runner until this one produces useful work you can inspect and measure. *** ## What's next? Turn the version that worked into a Workflow with its own slug. Give the job a schedule, sources, rules, and tool permissions.