Chidori is a reactive runtime and orchestrator for building durable AI agents, targeting developers who need to manage complex AI workflows. It provides solutions for understanding agent behavior, pausing and resuming execution, and managing state-space exploration, enabling developers to leverage familiar Python and JavaScript patterns.

How It Works

Chidori utilizes a Rust-based runtime that supports Python and JavaScript code execution, allowing for seamless integration of AI models into long-running workflows. Its core innovation lies in its reactive nature, enabling features like caching, state resumption, time-travel debugging, and visual debugging of execution graphs. This approach aims to reduce accidental complexity in AI agent development by providing robust tooling for state management and observability.

Quick Start & Requirements

• Installation: cargo +nightly install chidori-debugger --locked
• Prerequisites: Rust toolchain (nightly channel), Xcode command-line tools, CMake, Python 3.12+, uv.
• Runtime Setup: Requires LiteLLM proxy running at http://localhost:4000 and OPENAI_API_KEY environment variable set. Install LiteLLM with uv pip install "litellm[proxy]".
• Resources: Building dependencies may require significant compilation time.
• Docs: https://github.com/ThousandBirdsInc/chidori

Highlighted Details

• Runtime written in Rust, supporting Python and JavaScript.
• Time-travel debugging and branching for state exploration.
• Visual debugging environment for state graphs.
• Code interpreter environments with future containerized support.

Maintenance & Community

• Early open-source release seeking collaborators.
• Community support via Discord: https://discord.gg/chidori

Licensing & Compatibility

• MIT License.
• Compatible with commercial and closed-source applications.

Limitations & Caveats

The project is in early release, with containerized node support and agent re-evaluation with feedback planned for future development. Safeguards against running untrusted code are still being developed.