jamjet: JamJet is an action-control plane for AI agents that enforces safety policy across hooks, guardrails, MCP gateways, SDKs, and custom runtimes. It blocks unsafe tool calls, requires human approval for risky actions, caps costs, and provides a durable audit trail. One policy file works everywhere, supporting frameworks like Claude Code, OpenAI Agents SDK, LangChain, and more.; martin-loop: MartinLoop is a governance layer for AI coding loops that enforces budget caps, verifier gates, rollback evidence, and audit trails. It prevents unbounded retries by stopping unsafe or uneconomical work before execution, providing visibility and accountability for autonomous coding agents.
Calling MCP servers or arbitrary tools
Bounding AI agent retry loops with budget and safety constraints