JamJet
An open-source safety and control layer for AI agents that blocks unsafe tool calls, pauses for human approval on risky actions, enforces per-agent cost budgets, and keeps an audit trail that can survive a crash.
JamJet sits underneath an existing agent framework, such as Claude Code, the OpenAI Agents SDK, MCP clients, LangChain, CrewAI, Google ADK, or Spring AI, and enforces a single safety policy across all of them: blocking unsafe tool calls like database deletes or payments, pausing for durable human approval on risky actions, capping cost per agent, run, or project, recording an audit trail, and replaying or resuming a crashed run from its last checkpoint.
On the JVM, a Spring Boot starter auto-injects JamJet's advisors into a Spring AI ChatClient with no application code changes, adding durability and auditing to every call, and an embedded Java runtime avoids a network round trip to a sidecar for teams using Spring AI, LangChain4j, or Google ADK. A pip-installable Python package offers offline demos of the blocking, approval, and budget-cap behavior with no API key or cloud account required.