Chat with any Git Repo

It is a code-aware chat that loads an entire codebase and lets you ask questions about it in plain English. Paste a public GitHub repo URL (or upload your files) and the AI reads the project, then answers things like "where is auth handled?", "explain this function", "why does this test fail?", or "what calls this API?" — with references to the actual files. Think of it as pair-programming with someone who has already read your whole repo.

Two ways. (1) Paste a repo URL (e.g. https://github.com/owner/project) and click Load — we clone it and index the source. (2) Drag-and-drop your own files or a zipped folder if the code is private or local. Once it is indexed, just start chatting. Switch between loaded repos without losing the conversation.

Public GitHub repositories work today via URL, plus direct file/folder/zip upload for anything not on GitHub. GitLab and Bitbucket URL import are on the roadmap. Until then, for a GitLab or Bitbucket project just download the repo as a zip and upload it — the chat treats uploaded code identically to a cloned repo.

Yes. It runs on our self-hosted Qwen 3 Coder model on the daily free token pool — 2,500 tokens/day anonymous, 10,000/day with a free account. A typical question-and-answer turn costs ~100-300 tokens. Indexing a repo is cheap; you only spend tokens when you actually ask questions. Premium models (Claude Sonnet / GPT-5) are available per-call for very large or tricky reasoning over the codebase.

No training, ever. Uploaded files and cloned repos are held only for the duration of your session so the AI can answer questions, then cleared. Public repos you point at are already public. Private code you upload is never shared, never used to train models, and is deleted automatically. Full details at /privacy/.

Single files up to 10MB and zipped projects up to roughly 100MB / a few thousand source files index reliably. Very large monorepos are supported by indexing the most relevant files first — the AI retrieves the chunks tied to your question rather than stuffing the whole tree into one prompt. Binary assets, node_modules, build artifacts, and lockfiles are skipped automatically so the index stays focused on real source.

It indexes the whole repo, then retrieves the most relevant chunks for each question (retrieval-augmented). That keeps answers grounded in the right files without exceeding the model context window. When you ask about a specific function or file, it pulls that code plus its callers/callees; when you ask a broad architecture question, it pulls the structural overview. You can also @-mention a path to force it to focus there.

All the popular ones — Python, JavaScript/TypeScript, Java, C/C++, C#, Go, Rust, Ruby, PHP, Swift, Kotlin, plus config (YAML, JSON, TOML), SQL, shell, Dockerfiles, and framework conventions (Django, Flask, React, Vue, Spring, Rails, Express, and more). Because it is the same Qwen 3 Coder engine behind our /code/ generators, it recognizes idioms and project layout, not just syntax.

Yes — ask it to fix a bug, refactor a function, add error handling, or implement a feature and it returns the modified code with an explanation of the change. It is read-only against your repo (it never pushes commits), so you copy the suggested diff and apply it yourself. For a dedicated fix-and-debug flow use /code/debug/, and /code/refactor/ for larger restructures.

Yes. Point it at a module and ask for unit tests (pytest, Jest, JUnit, etc.) and it writes them against the real signatures it read. Ask for a README, docstrings, or API docs and it generates them from the actual code. For one-shot versions outside a repo chat, /code/test/ and /code/documentation/ and /code/readme/ are the dedicated tools.

Plain /chat/ answers from general knowledge and whatever you paste into the message. Chat with Repo grounds every answer in your actual indexed codebase — it cites real files, follows imports across modules, and stays consistent with your project conventions. Use /chat/ for general coding help; use this when the answer depends on understanding how your specific project is wired together.

Yes — the underlying code chat is reachable via POST to /v1/chat/ on api.free.ai with Bearer auth using a developer API key. Send your repo context or file contents as part of the request and stream the answer back, OpenAI-compatible. Useful for building code-review bots or CI assistants. Python / Node / cURL snippets at /api/.