Hey Product Hunt, I'm Wahib, the maker. Most agent tooling helps you build a workflow. My problem started after that: the workflow had to run as a long-lived process, fail clearly, restart carefully, and stay inspectable afterwards. Debugging step 7 of a run meant paying for real API calls again, and never getting the same behaviour twice. So AgentOS journals every LLM exchange and tool result at the provider boundary. `agentOS replay` re-runs a session offline with the recorded responses and reports drift. One process also gives you supervision, a health endpoint, a gRPC bus, an SSE event stream and secret isolation via a vault crate. It is built in Rust and designed to sit under LangGraph, AutoGen, CrewAI or your own agents rather than replace them. Being honest about the stage: the run / ps / logs / trace / replay CLI flows are stable locally. `agentOS fork` (branching a run from a checkpoint) is scaffolded but not implemented yet — that's what I'm building next, along with the dashboard, WASM plugin runtime and stronger restart guarantees. One question I would love answers to: if you run agents today, what breaks first for you — supervision, observability, or reproducing failures?

AgentOS

Rust runtime for AI agents with time-travel replay

Rust runtime for AI agents with time-travel replay