A C++ code reviewer that runs in your terminal. Scans your project, sends code to an AI provider, and highlights what needs fixing -- without giving away the answer.
cargo install carbon-reviewOr build from source:
git clone https://github.com/BABYB0I/carbon.git
cd carbon
cargo build --release
./target/release/carbon-review --helpRun it and follow the prompts:
carbon-review ./srcIf no API key is configured, you'll be asked for your provider and key on first run. Press s to change settings later.
carbon-review [OPTIONS] [DIR]
--provider Provider (groq, ollama, openai)
--model Model name (overrides env)
--summary Batch mode -- generate a report
--format Output format: html, md, txt (default: html)
--output Output file path
--delay Seconds between API calls (default: 2)
| Provider | Env Variable | Default Model |
|---|---|---|
| Groq | GROQ_API_KEY | llama-3.3-70b-versatile |
| OpenAI | OPENAI_API_KEY | gpt-4o-mini |
| Ollama | OLLAMA_URL (optional) | llama3.2 |
| Key | Action |
|---|---|
| Up / Down | Navigate file tree |
| j / k | Scroll review (after review) or navigate tree |
| Enter | Review file / Force re-review |
| Left | Collapse directory |
| s | Open settings (provider, model, API key) |
| r | Rescan directory |
| Ctrl+j / Ctrl+k | Scroll code panel |
| q / Esc | Quit |
Carbon walks the directory tree looking for C/C++ files (.c, .cpp, .cxx, .cc, .h, .hpp, .hxx). Each file is sent to an AI provider for review. The AI identifies bugs, undefined behavior, and readability issues, providing hints without fixing the code for you.
Reviews are cached by line content. If you edit a file and re-review, only changed blocks are sent to the AI — unchanged lines keep their previous critique. Responses stream token-by-token with a live timer and extracted score.
MIT