FAQ

What is Choreographr?

Choreographr is an all-purpose, extensible AI agent system written entirely in Rust. It has a client/server architecture and can run many sessions simultaneously. It runs locally or in the cloud, and LLM-generated code can be executed in a sandboxed RISC-V VM for complete security and observability.

What can I use it for?

Software development, a personal/business agent, or research. It runs on your desktop or in the cloud, and you can talk to it from the terminal, a desktop app, or your phone (via the Telegram bridge).

Which models does it support?

70+ providers across three wire protocols: OpenAI-compatible (OpenAI, DeepSeek, Mistral, xAI, Groq, Ollama, OpenRouter, and many more), Anthropic Messages (Claude and friends), and Google Gemini.

Is my data private?

Choreographr is local-first. Sessions and credentials live on your machine (encrypted at rest), and you choose which model providers your prompts go to. The daemon starts locked and credentials are only decrypted into memory after unlock.

Is LLM-generated code safe to run?

Generated code runs in an isolated RISC-V VM by default, which is a complete replacement for the shell tool — full control and observability. OS-level sandboxing for tools that must run outside the VM is on the roadmap.

Does it require an async runtime?

No — the daemon uses pure OS threads with message passing. No async code in its own logic, no Arc<Mutex> shared state. A tokio runtime exists only as a sidecar for third-party crates that require it.

Is it open source?

Yes — Apache-2.0.

What's the minimum supported Rust version?

1.91.

How is Choreographr different from other agents?

Its daemon + multi-client architecture, sandboxed RISC-V VM, subsessions (hierarchical subagents), and per-session undo/redo are a distinctive combination. See the comparison on the homepage.