> ## Documentation Index > Fetch the complete documentation index at: https://docs.coverbase.com/llms.txt > Use this file to discover all available pages before exploring further. # Working with Coverbase AI > Start a Coverbase AI run, choose whether to watch it or run it in the background, answer its questions and approvals, steer it while it works, and read the report and to-dos it leaves you. For AI agents: a documentation index is available at [https://docs.coverbase.com/llms.txt](https://docs.coverbase.com/llms.txt). This page is also available in markdown by appending .md to the URL. > [!NOTE] > > This guide is part of the [User Guides](https://docs.coverbase.com/user-guides/overview.md) collection. It covers **Coverbase AI** in the left navigation and the Coverbase AI button in the top bar. It sits beside [Files and records](https://docs.coverbase.com/user-guides/coverbase-ai-files-and-records.md) and [Memory, permissions and standing agents](https://docs.coverbase.com/user-guides/coverbase-ai-memory-and-agents.md). For what Coverbase AI is, see [Coverbase AI](https://docs.coverbase.com/products/coverbase-ai.md). > [!NOTE] > > Your Coverbase representative turns on Coverbase AI for your organization. Coverbase AI does a job for you and reports back. You describe the job in your own words, and it works through the product with your permissions. The mistake people make most often is describing a job without saying which records it is about. "Review the SOC 2 report" leaves it to search. Start from the vendor's page, or pick the vendor and attach the report from the **+** menu, and it starts on the right record with the right file. ## Step 1: Open Coverbase AI Click **Coverbase AI** in the left navigation, or the Coverbase AI button in the top bar. The top bar button is on every page and shows what your latest run is doing: working, waiting on an assessment, needing you, or done with a count of to-dos. Hover it to see the plan, approve what it is waiting on, or start another run. When you open Coverbase AI from a record's page, such as a vendor, an assessment, one control's evaluation, a contract or a Radar signal, the composer names that record. "This", "it" and "here" in your instruction mean that record. ## Step 2: Describe the job Type what you want done and press **Enter**. **Shift+Enter** adds a new line. Some examples: | You type | It does | | - | - | | Onboard Figma for design collaboration | Researches the company, opens the intake, matches the vendor, runs screening, drafts and submits the inherent risk questionnaire, decides whether an assessment is needed and launches it | | Review this assessment | Goes through the controls that need a person, proposes an outcome for each with its reason, and raises findings and obligations you approve | | Watch our critical vendors on Radar | Sets up detectors for the matching vendors, left off until you approve turning them on | | What changed for Okta since last quarter? | Answers from the records and, where the question reaches outside Coverbase, the web, with a link or source for every claim | | Build me a findings dashboard | Adds a dashboard from the template that answers the question | Pick **Playbooks** on the home page to start from a ready-made instruction. Before you start, choose how the run works: * **Watch it work**: the browser follows the agent through the product. A cursor moves to each field it fills and a caption says what it is doing and why. * **Run in background**: nothing moves on your screen. You are notified when it needs you and when it is done. Closing the tab turns a watched run into a background run. The run itself never stops because you left. * **Don't stop to ask**: approves every step it would otherwise ask about, as long as the step can be undone. Archiving a record, submitting an intake request and anything else that cannot be undone still ask. ## Step 3: Answer what it needs When a run needs you, it stops and says why at the top of the run page. The top bar button shows **Needs you**, and you get a notification. | It shows | What to do | | - | - | | A question with numbered options | Pick an option (press **1** to **5**), or type your own answer. The option it would pick is marked **Recommended**. | | An approval, such as launching an assessment or turning a detector on | Click the approve button to go ahead, or the other option to keep things as they are. Typing an answer instead tells it what to do. | | A list of proposed changes, such as control outcomes from a review | Untick the ones you disagree with and click **Apply**, or **Keep all as they are**. | | The inherent risk review | Agree or disagree with each risk domain, move a level and say why, tick the assessments to run, and confirm. | | A request for a file | Drop the file on the card or click **choose files**, then **Send file**. See [Files and records](https://docs.coverbase.com/user-guides/coverbase-ai-files-and-records.md#when-coverbase-ai-asks-for-a-file). | An archive card says exactly what goes with the record, such as "Its 3 assessments, 12 documents and 2 obligations go with it", and its button carries the verb. You can also answer from outside the app. A needs-you email has one button per option, and you can reply to it in your own words. If your organization uses Slack or Microsoft Teams with Coverbase, approvals can come to you there. A batch of changes and the inherent risk review are answered in the app. ## Step 4: Steer it while it works The box under the run says **Tell Coverbase AI to change course**. What you type depends on what the run is doing: | The run is | Your message | | - | - | | Working | Is read at its next step. It says how it will change course and rewrites the steps still to do. | | Asking a question | Counts as your answer. | | Waiting for an approval | Withdraws the question. Nothing is done until it reads your message. | | Waiting on an assessment | Wakes it now instead of waiting. | | Done | Picks the run back up with everything it already knows. | Click **Stop** to end a run at its next step. Nothing more is written after that. ## Step 5: Read the report Every run ends in a report. It says what it did, with a link to each record, what you need to do, and what it could not do and why. A run that answered a question shows the answer. ![Report of a run that read five attached files, listing why each could not be read, followed by What you need to do with one to-do per file](https://mintcdn.com/coverbase/lFilVcpKbisiv4nI/images/user-guides/coverbase-ai-report.png?fit=max\&auto=format\&n=lFilVcpKbisiv4nI\&q=85\&s=4aa60dc949cbd3db7466e559eeca65ec) A finished run: the answer with its sources, and the to-dos it left you. * **What you need to do** lists the to-dos. When the run is about one vendor, each becomes a [work queue](https://docs.coverbase.com/user-guides/analyst-reviewer.md) item assigned to you. Finishing the item ticks the to-do on the report. * **What cannot be undone** lists every action the run took that has no way back. Read it first after a background run. * **Because you told me** lists the remembered preferences that changed what it did. See [Memory, permissions and standing agents](https://docs.coverbase.com/user-guides/coverbase-ai-memory-and-agents.md). * **Copy report** copies the whole report as text. **Export** downloads the run, step by step. ## Keyboard shortcuts Single keys work while no field has focus. Press **?** on a Coverbase AI page to see them all. | Key | Does | | - | - | | **/** | Focus the composer | | **Enter** / **Shift+Enter** | Send / new line | | **A** | Approve, or apply the ticked changes | | **K** | Not now, or keep everything as it is | | **1** to **5** | Pick an option on a question | | **Esc** | Back to all runs, then back to the page you came from | **A** and **K** do not archive or submit anything. Those take a click. ## Troubleshooting | Problem | What to do | | - | - | | There is no Coverbase AI in the navigation | It is not on for your organization yet. Ask your Coverbase representative. | | The run says it cannot do something | You may not have the permission the step needs. The run acts as you and reports a refusal rather than working around it. Ask an admin, or have someone with access run it. | | The run says a module is not covered | RFPs, Inspect, risk scenarios, sending a contract for signature and allowlisting a screening match are not available to Coverbase AI yet. It says so and does the rest of the job. | | The run is waiting on an assessment | It resumes on its own when the assessment finishes. Type a message to wake it sooner. | | It picked the wrong record | Tell it which one in the box under the run, or start again from the record's page. | | It ran something you did not want | Look it up in the [Agent Ledger](https://docs.coverbase.com/user-guides/agent-ledger.md), which shows every record the run changed and who approved it. | ## Related > **[Files and records](https://docs.coverbase.com/user-guides/coverbase-ai-files-and-records.md)** > > Attach documents and point a run at the right records. > **[Memory, permissions and standing agents](https://docs.coverbase.com/user-guides/coverbase-ai-memory-and-agents.md)** > > Preferences, per-action permissions and agents that run on their own. > **[Agent Ledger](https://docs.coverbase.com/user-guides/agent-ledger.md)** > > Every action an agent took, and who decided. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.