What it does ChatPilot is a full-stack chat interface that wires together common LLM agent patterns: Google search via Serper or DuckDuckGo, file/URL ingestion for RAG, and a Python code interpreter that can run in either E2B’s sandbox or a local environment. It speaks to OpenAI, Azure, Ollama, or anything LiteLLM supports. The frontend is Svelte; the backend is FastAPI.

The interesting bit The project explicitly positions itself as a clone of Kimi Chat’s “drag files in, paste URLs out” workflow. What’s slightly unusual is bundling both a remote sandbox (E2B) and a local Python executor as options — most projects pick one and call it done.

Key highlights

• Multi-model backend: OpenAI/Azure, Ollama, or LiteLLM config file for cloud providers
• RAG pipeline for uploaded files and URL content extraction
• Voice I/O and image generation support (mentioned, not detailed)
• User management with permission controls and chat export/import
• One-liner Docker run with env-based configuration

Caveats

• The README notes the code is “还很粗糙” (still quite rough) and asks contributors to add unit tests
• Frontend builds require Node ≥ 20.10, or you can unzip a pre-built build.zip — the latter feels like a workaround
• Voice and image generation features are listed but not explained; it’s unclear how they integrate

Verdict Worth a look if you want a self-hosted, agent-capable chat UI without building the plumbing yourself. Skip it if you need polished UX or deep customization — the author is upfront that it’s early-stage.