jamjet-labs/jamjet

View on GitHub ↗

The open-source safety layer for AI agents — block unsafe tool calls, require approval, enforce budgets, audit, replay.

20 ★5 forksRustUpdated 1mo ago

What you need to know

Action-control plane for AI agents that enforces policies, blocks unsafe tool calls, pauses for human approval, caps costs, and records audit trails across hooks, guardrails, MCP gateways, and SDKs.

Install

pip install jamjet
jamjet demo unsafe-tool-call

Usage

  • Drop a policy.yaml in ~/.jamjet and run the runtime or use adapters for Claude Code, OpenAI Agents SDK, MCP clients, LangChain, Spring AI

Key features

  • Block unsafe tool calls at runtime
  • Durable pause for human approval
  • Per-agent and per-run cost caps
  • Audit trail with signed export
  • Crash recovery and replay from checkpoints
  • Adapters for Claude Code, OpenAI Agents SDK, MCP clients, LangChain, CrewAI, ADK, Spring AI
  • Spring Boot starter for Java
  • Engram memory layer (MCP, Python, Java)
  • Durable, event-sourced execution engine

Best for

Adding runtime safety and governance to AI agents

Caveats

  • Rust runtime with Python, Java, and TypeScript SDKs
  • Apache-2.0
Platforms: Python · Java · Node.js · RustClients: Claude Code · OpenAI Agents SDK · MCP clients · LangChain · CrewAI · Spring AI
Documentation ↗

Reviewed 2026-08-11

Topics

a2aagent-memoryagent-safetyagentic-aiai-agentsai-governanceai-safetyapproval-workflowsaudit-loghuman-in-the-loopjavallmllmopsmcpmodel-context-protocolproduction-aipythonrustworkflow
Stars
20★
Forks
5
Language
Rust
License
Apache-2.0
Created
2026-03-07
Last push
2026-08-03