FAQ
AI coding agent questions, answered
Control, security, languages, accuracy, and what happens to your code. The honest answers.
No. Agentcode opens a pull request; you review and merge. The agent never merges on its own. Its unit of work is a reviewable diff with a written plan and a passing test run, delivered into the pull request flow your team already trusts. Nothing reaches your main branch until a person approves it, so adopting an autonomous agent does not mean giving up control of what ships.
An AI coding agent is software that completes a whole coding task end to end, not just a single suggestion. You describe a feature or a bug, and the agent plans the work, edits multiple files across your codebase, runs the test suite, and opens a pull request for review. It is the layer above autocomplete and inline chat: instead of suggesting the next line while you type, it does the task and hands you a reviewable result.
Copilot and Cursor assist you inside the editor with autocomplete and inline chat while you write code. Agentcode works asynchronously on your repository and delivers a finished pull request with passing tests. You do not have to babysit a chat window or learn a new IDE; you describe the task, the agent runs, and a reviewable PR shows up in GitHub or GitLab. It complements an in-editor assistant rather than replacing your editor.
Agentcode is review-first and never trains on your code. Your repositories, prompts, and diffs are not used to train any model, ours or anyone else's. Code is encrypted in transit and at rest, access is least-privilege and scoped to the repos you connect, and we operate under SOC 2-track controls. Because the agent only ever opens a pull request, you see exactly what it changed before anything is merged.
Agentcode works with Python, TypeScript, JavaScript, and Go today, plus the common frameworks and test runners around them, including PyTest, Jest, Vitest, and Playwright. It connects to your existing GitHub or GitLab repository and runs your real test and lint commands, so it follows your stack and your CI rather than asking you to migrate to a new one.
Agentcode will not open a pull request until the test suite is green. If it cannot get the tests passing, it tells you what blocked it instead of shipping broken code. And because the result is always a pull request, the worst case is a PR you read and close. You lose nothing: you never have unreviewed code on your main branch, and you only merge what you are happy with.
No. Agentcode runs on your repository through GitHub or GitLab, so there is nothing to install in your editor and no new IDE to learn. You keep your current editor, your current workflow, and your current CI. The agent simply opens pull requests you review like any other.
You describe a task in plain language. The agent reads the relevant parts of your codebase, writes a short numbered plan, edits the files it needs, and runs your test suite. When the tests pass, it opens a pull request that shows the plan, the diff, and the test run on a branch named for the task. You review the PR, request changes if needed, and merge when you are ready.
Plans start at $29 per month billed yearly for solo developers, with Pro at $59 per month for power users and small teams, and Team at $129 per month for growing engineering teams. Every plan is paid; there is no free tier. You can try the interactive demo on this page for free to see exactly how a task becomes a reviewable pull request before you sign up.