# VlinderCLI **Debug and repair AI agents.** --- ## What is VlinderCLI? Infrastructure to debug and repair AI agent failures. Rewind to the failure, inspect the exact state, test a fix, and replay. --- ## Core Concepts - **Agents** — self-contained AI units running in OCI containers - **Fleets** — groups of cooperating agents with delegation - **Sessions** — independent Merkle chains with content-addressed history - **Time-Travel Debugging** — rewind, fork, fix, promote --- ## Architecture Queue-based, protocol-first design. Every component communicates through typed messages over NATS. -v- ### Workers | Worker | Role | |--------|------| | Registry | Source of truth (gRPC) | | Agent | OCI container execution | | Inference | LLM text generation | | Storage | Object + vector | -v- ### Message Flow 1. **Invoke** — CLI sends user input 2. **Request** — agent calls a service 3. **Response** — service returns result 4. **Complete** — agent finishes --- ## Time-Travel Debugging ```text $ vlinder session list todoapp $ vlinder session fork ses-abc --from a1b2c3d4 --name fix $ vlinder agent run todoapp --branch fix $ vlinder session promote fix ``` Rewind to any completed turn. Fork. Fix. Promote. --- ## Get Started - [Documentation](https://docs.vlindercli.dev/) - [GitHub](https://github.com/vlindercli/vlindercli) - [Blog](https://blog.vlindercli.dev/)