Chidori

An open-source AI agent framework for macOS and Linux with disk checkpoints, recorded-run replay and support for Ollama and OpenAI-compatible backends.

Chidori is an AI agent framework for developers building long-running workflows that need crash recovery or human approval. Agents use ordinary async TypeScript, with saved runs that developers can revisit for debugging and testing.

Recorded-run replay Model responses, tool results and HTTP calls are recorded so developers can replay a run using the same code without making fresh model requests.
Crash recovery Disk checkpoints let an interrupted agent continue in a new process after a crash or restart.
Approval pauses An agent can save its state while waiting for a person or another agent to respond, without keeping a process running during the wait.
Regression checks Recorded checkpoints can serve as integration-test fixtures for comparing agent behavior without repeating live model calls.
Provider caching Chidori marks stable prompt prefixes for caching by providers that support it.

The self-contained Rust runtime runs on macOS, including Apple Silicon and Intel machines, and Linux. It includes a JavaScript engine and doesn't require Node or Deno. TypeScript and Python SDKs connect to the runtime over HTTP.

Chidori supports OpenAI-compatible chat-completions backends, including Ollama, vLLM, LiteLLM, DeepSeek and Groq. The runtime and checkpoints stay on your machine or server; live model requests go to the backend you choose, which may be local or hosted. Replay depends on the recorded call log and the same agent code. The project is open source under Apache 2.0.