Skip to main content

FAQ

What is CheepCode?​

CheepCode is an AI coding agent that works in your ticketing system (Linear only for now, with support for Jira, Trello, Monday, Asana, and more coming very soon).

How does it work?​

CheepCode uses a combination of LLMs and specialized tooling to write code for you. When you assign a task to CheepCode, an agent will clone your repository, do its best to complete the task, and submit a pull request.

Is my code ever stored on CheepCode's servers?​

CheepCode downloads your repo fresh each time it runs, and the ephemeral container with your code is destroyed after the task is complete.

How does CheepCode know what tasks to work on?​

CheepCode will work on any task that is:

  1. on a configured ticket board (in a "Project" in Linear that you've linked with a CheepCode project), and
  2. in your ticketing system's "ready" status (e.g. "Todo" in Linear).

Our agent is constantly checking for new tasks, and starts work on new items within minutes.

How long does it take?​

CheepCode is able to complete many tasks in a matter of minutes once it starts work. The agent will work for up to 30 minutes on a given task.

What if the agent doesn't complete the task?​

In general, if you get a result that isn't an error message, the agent has "completed" the task and you've been charged for the run.

Below, we'll unpack a few ways that the agent's output might fail to meet expectations, and what that means for billing:

  1. The agent submits a result that misunderstands the requirements.

    This is a common case, even for human developers!

    You will still be charged 1 credit for the run, because the agent attempted the task.

    When this happens, we recommend revising the ticket with more context before putting the ticket back into the "ready" state (e.g. "Todo" in Linear) for another run.

  2. The agent runs out of time working on the task.

    This is a rarer case for CheepCode, and is more likely to happen if the task is very complex. The agent will still submit a PR in this case, but will also leave a comment explaining which parts it was able to complete and which parts remain.

    You will still be charged 1 credit for the run, because the agent attempted the task.

  3. The agent starts working on the task, and realizes it doesn't have enough information to complete the task.

    This is another common occurrence when working with AI or human engineers. In this case, the agent will submit a comment on the ticket explaining what it was able to figure out and what it needs to clarify.

    Because the agent attempted the task and provided a result, you will be charged for the clarifying run.

    Stay tuned for updates here! We're working on incorporating this process into a first-class feature.

  4. The agent can't clone your repo, crashes mid-run, or is otherwise unable to submit results.

    Failures do happen! In this case, you will not be charged for the failed run. We'll put the ticket back into the "ready" state (e.g. "Todo" in Linear) and it will be re-tried. If the agent encounters continued failures, we will put the ticket into the "pre-ready" state (e.g. "Backlog" in Linear) and stop trying.

How can I teach the agent about my codebase?​

CheepCode can read any documentation in your repository, including Cursor Rules and other AI guidance files. One neat thing about modern AI agents is that any documentation that benefits them is likely to benefit you, too. A well-documented codebase is a happy codebase, whether it's humans or AI working in it.

How does the agent install the tooling it needs?​

The agent runs your code in a container, and your repository must contain a Dockerfile that the agent can use to build that container and run your code.

Can the agent run my unit/integration/end-to-end tests?​

Yes! If your project clearly documents your test command (e.g. bun test), and your container is configured correctly, then the agent will be able to run your tests.

How many tasks can the agent run at once?​

While CheepCode is in private beta, we aren't strictly limiting the number of tasks that can be run at once. In practice, you can expect up to a dozen tasks running at a time, depending on system load.

How do I get support?​

If you have any questions or feedback, please email us. Support is offered on a best-effort basis, with no SLA or guaranteed response time.