One task. A clear path through the code.

A working first step: repository planning.

The browser preview inspects the project files you select, identifies languages and package scripts, and ranks files against your task description. It creates a development outline grounded in those files, with checks and review questions you can export as Markdown.

Analysis happens locally. There is no account, API key, cloud upload, model call, code editing, or test execution in this preview.

The planned agent workflow.

The next product direction extends planning into implementation and verification. This concept shows how the full workflow could fit together; it is not a recording of a completed agent run.

DevPilotWorkflow concept

EXAMPLE REPOSITORY / billing-service

Add rate limiting to the public API.

01   Plan02   Changes03   Review

A plan before a patch

1. Find the request path and existing middleware.src/server.ts · src/routes/v1/
2. Implement a per-key rate limiter.src/middleware/rate_limit.ts
3. Cover allowed and rate-limited requests.test/rate_limit.test.ts
4. Run checks and prepare a review summary.npm test · git diff

Context → plan → patch → checksYou decide what ships

Concept for the planned agent workflow. To try the working local planning tool, open the planning preview.

1. Gather the context

Read relevant modules and development instructions. Identify the behavior that needs to change and the checks that can help verify it.

2. Make the plan explicit

Map the request to a sequence of changes. Keep the scope visible so the developer can assess the approach.

3. Implement and inspect

Apply the work in a separate Git branch. Keep the resulting diff connected to the plan and original request.

4. Check and hand back

Run relevant checks, capture their actual results, and explain the changes. Surface failures and open questions in the handoff.

Make verification visible.

Passing tests provide evidence, but they do not replace review. A useful handoff should include what changed, which checks ran, what failed, and anything that remains uncertain.

• A Git diff tied to the requested task.
• Test commands and their actual output.
• Skipped checks and unresolved issues.
• A developer's decision about merging and deployment.

Claude for reasoning. DevPilot for the workflow.

The planned Claude API integration will provide reasoning, task planning, and code generation. DevPilot will coordinate repository context and development tools around the model.

The intended loop is to read context, choose the next step, call a development tool, inspect its output, and update the task. The model reasons about the output; the toolchain performs file, test, and Git operations.

Where development stands.

1. AVAILABLE NOWLocal repository-planning previewProject-file analysis, package-script discovery, task-related file suggestions, and Markdown plan export.
2. NEXTClaude integration and agent toolingConnect planning to model reasoning, multi-file implementation, test execution, and Git handoff.
3. LATERPrivate testing of the full agentValidate behavior on real projects, refine the review experience, and document setup and product data handling.

A full-agent release date has not been announced. Contact the founder to discuss your use case.