JamJet Documentation
The agent-native runtime — durable execution, native MCP + A2A, full observability. Built in Rust. Authored in Python.
JamJet Documentation
Welcome to JamJet — the production-grade runtime for AI agents.
What is JamJet?
JamJet is an open-source, durable workflow runtime built specifically for AI agents. The core is Rust (for performance). The authoring surface is Python and Java (for ergonomics). Every execution is event-sourced, checkpointed, and replayable.
Quick links
- Quickstart — Get a durable agent running in under 5 minutes
- Concepts — Understand the architecture
- Python SDK —
@task,Agent,WorkflowAPI reference - Research Guide — Run reproducible multi-agent experiments
- MCP Guide — Connect to any MCP tool server
- A2A Guide — Agent-to-agent communication
Install
pip install jamjetKey capabilities
| Capability | Description |
|---|---|
| Durable execution | Event-sourced, crash-recoverable, checkpoint replay |
| Native MCP + A2A | Client and server for both protocols |
| Six strategies | react, plan-and-execute, critic, reflection, consensus, debate |
| Built-in eval | LLM-judge, assertions, cost/latency scorers, custom @scorer |
| ExperimentGrid | Parameter sweeps with LaTeX/CSV/JSON export |
| Rust core | Async scheduler, microsecond dispatch, real parallelism |
| Polyglot SDKs | Python, Java, Go (coming) |
| Policy engine | Token/cost limits, PII redaction, autonomy levels |