About CriogenAI

CriogenAI is an open-source AI agent framework. A Rust engine runs agents (a model with tools, in a loop) and workflow graphs, and you write them in Python, TypeScript or Rust. It is built to be the fastest way to run AI agents in production: in our benchmark, an agent turn costs 0.29 ms of framework overhead, and 1,000 concurrent agent runs finish in a third of a second.

Why we build it

Agents that do real work crash halfway through tasks, call tools that move money or change records, and, in regulated fields like healthcare and finance, must account for every step. We think those problems belong in the runtime, not in every team's own glue code. So CriogenAI saves every step and resumes after a crash without repeating finished work, stops for human approval before sensitive actions, encrypts data at rest and keeps a tamper-evident audit trail. Models can run in the cloud or on your own servers, so sensitive data can stay inside your network.

What it includes

• Agents with tools, structured output, retries and fallback models
• Workflows: graphs with parallel branches, joins, loops and human-in-the-loop pauses
• Durable threads on SQLite or Postgres, memory with keyword and semantic search
• MCP tools, evals, tracing, collections and retrieval
• An LLM gateway with OpenAI and Anthropic wire formats
• Over 15 model providers, including OpenAI, Anthropic, Gemini, Azure, Bedrock, Vertex AI, Ollama and vLLM

Key facts

Contact

Questions, bugs and ideas: GitHub issues. Security reports: see the security policy. News: the blog and its RSS feed.