Chidori
A reactive runtime for building durable AI agents
Last verified:
What is Chidori?
Chidori is an open‑source framework for building agent systems that orchestrate generative AI with traditional software and web services, emphasizing observability, replay, and deterministic execution. It lets developers define workflows in Starlark, a deterministic dialect of Python, and run them inside a Rust agent runtime that provides a reactive database, simple wrappers for common agent operations, and SDKs for multiple languages. Core use cases include chatbots, retrieval‑augmented generation systems, integrations with external services, and fully autonomous agents that can interact with the world over time.
The framework centers on structured debugging and visualization: users can inspect an agent’s internal state, trace execution steps, and use time‑travel debugging to replay and analyze past runs. Chidori also supports composition and triggers, enabling developers to assemble complex agent graphs from smaller primitives and react to events such as external API changes or user messages. The Thousand Birds ecosystem provides a hosted environment and tooling so teams can ship agent systems they can understand, operate, and deliberately improve.
Chidori is aimed at developers and product teams building AI‑driven applications where reliability, security, and debuggability are critical. It is especially useful for teams that want fine‑grained control over agent behavior, audit trails of agent decisions, and cross‑language interoperability (for example, different parts of an agent system in Python and JavaScript). The framework is designed to sit below higher‑level products, offering a programmatic layer for orchestrating agents rather than a no‑code UI, making it well‑suited for engineers comfortable with code and infrastructure rather than general end users.
Key patterns include RAG‑style agents enhanced with your own data, chat fronts that route messages through Chidori workflows, and autonomous agents that invoke external tools or APIs conditionally. The reactive model and deterministic runtime allow for repeatable behavior across runs, which helps testing and continuous improvement. By focusing on simplicity, performance, and security, Chidori positions itself as a lightweight alternative to heavier, more generic agent frameworks that bundle many abstractions and are harder to inspect and debug at runtime.
Chidori pricing
Pricing model: Freemium
Chidori is open source and available on GitHub, so the core framework itself is free to use and self‑host. Thousand Birds provides a hosted environment and tooling around Chidori, but the docs and public marketing pages do not list explicit paid plans, tiers, or free credits; instead they emphasize using the open‑source framework and self‑hosted Debugger instance. For any commercial or managed usage, pricing is custom and must be negotiated directly with the Thousand Birds team via sales contact.
Chidori pros
- Open‑source core framework
- Rust‑based agent runtime for performance
- Deterministic execution and replay
- Time‑travel debugging for agents
- Reactive database for agent state
- Starlark‑based workflow definition language
- Built‑in primitives for common agent operations
- SDKs in multiple programming languages
- Strong focus on observability and tracing
- Visualizations of agent logic and execution
- Support for autonomous agents that interact with external systems
- Integration guides for RAG and chatbots
- Trigger‑based reactivity to external events
- Easily inspect and manipulate internal agent state
- Community‑driven development and contributions
Chidori cons
- Requires Rust and Starlark knowledge
- Steep learning curve for non‑developers
- Limited no‑code or low‑code tooling
- Fewer prebuilt UI templates than full platforms
- Sparse prebuilt connectors to niche SaaS tools
- Small company size and ecosystem compared to major vendors
- Limited documentation examples for complex multi‑agent setups
- No public SLA or enterprise‑grade support tiers mentioned
Frequently asked questions about Chidori
What type of applications is Chidori best suited for?
Chidori is best suited for building agent systems that combine generative AI with traditional software and web services, such as chatbots, retrieval‑augmented generation agents, integrations that react to external events, and autonomous agents that can take actions over time. It is designed for developers who want deterministic, observable, and replayable behavior rather than one‑off prompt‑and‑response scripts.
How does Chidori differ from LangChain or LlamaIndex?
Chidori focuses on a simpler, more performant, and secure agent runtime with built‑in observability, replay, and time‑travel debugging, whereas LangChain and LlamaIndex are more general‑purpose frameworks that bundle many integrations and abstractions. Chidori runs agents in a Rust runtime with a reactive database and deterministic Starlark workflows, aiming to reduce framework overhead and make debugging deeper and more systematic.
Is Chidori free to use?
Yes, Chidori is open source and available on GitHub, so you can self‑host and run the framework without licensing fees. Thousand Birds builds a hosted environment and tooling on top of Chidori, but detailed commercial pricing for those managed services is not published on the website and must be discussed directly with the team.
Which programming languages does Chidori support?
Chidori provides SDKs and client libraries for multiple languages, including Python and JavaScript, so developers can author and interact with agents in environments they already use. The core runtime is written in Rust, but the SDKs abstract that away so most agent logic can be expressed in Starlark or through the language‑specific bindings.
What are the core abstractions in Chidori?
Chidori’s core abstractions include code as the base primitive, composition to build complex agents from smaller pieces, triggers that react to events, an ecosystem layer for importing and exporting agents, time‑travel debugging to inspect past runs, and visualization tools to display an agent’s logic and execution flow. These abstractions are designed to make agent systems easier to understand, debug, and maintain over time.
Can I run Chidori entirely on my own infrastructure?
Yes, you can self‑host Chidori by installing an instance of the Chidori Debugger and running the open‑source framework on your own servers or containers. The framework is designed to integrate with your existing infrastructure so you retain control over security, latency, and data residency.
How does Chidori handle debugging and observability?
Chidori provides deep observability through time‑travel debugging, where you can replay agent runs step by step, inspect internal state, and visualize logic and execution paths as they unfold. The Debugger and visualization tools are built to help you trace why an agent made certain decisions and iterate on its behavior in a controlled way.
What is the role of Starlark in Chidori?
Starlark is used as the deterministic workflow definition language in Chidori, allowing developers to write agent logic that is easier to test, replay, and reason about compared to general Python. It provides a subset of Python‑like syntax without the unbounded side effects, which supports the framework’s focus on predictability and auditability.
Does Chidori support RAG‑style agents?
Yes, Chidori includes guides and patterns for building retrieval‑augmented generation agents that pull your own data into agent workflows. These agents can query external knowledge sources, apply retrieval logic, and then synthesize answers using generative models, all orchestrated within Chidori’s runtime.
Who is the target audience for Chidori?
Chidori targets developers and product teams that want to build reliable, observable AI agent systems, especially those who need deterministic behavior, audit trails, and the ability to debug and iterate on agent logic over time. It is particularly appealing to teams already comfortable with Rust, Starlark, and infra‑level tooling who want more control and less abstraction than generic agent frameworks provide.