Stain Lu builds open-source agent infrastructure, including OpenClaw Managed Agents and Hermes Labyrinth. His projects turn personal agent runtimes into programmable services and expose the evidence of what an agent did.[1][2][3]
Start with OpenClaw Managed Agents →
Links
Contributions
Expose a runtime through an application API
OpenClaw Managed Agents places a service layer around OpenClaw. Developers create an agent, open a session, send events and stream results through an API. Each active session uses an isolated Docker container; orchestrator metadata and durable state are stored separately so a session can survive the replacement of its live container. The project supplies SDKs and a deployment configuration for self-hosting.[2]
Trace an agent’s path through a task
Hermes Labyrinth is an observability plugin for Hermes Agent. It displays prompts, tool calls, failures, approvals and transitions between models or delegated sessions as an ordered journey. Its inspector links individual steps to inputs, outputs and timing, with redacted Markdown and JSON reports. The public demonstration uses mocked state; the installable plugin reads local Hermes records.[3]
Ideas
Labyrinth is designed to read state without starting, resuming or changing sessions. Its redaction path hides trace content if the core redactor cannot be loaded. That separation lets an operator inspect an agent’s work without granting the same interface authority to run it. The repository documents installation and verification of the plugin as well as the visual dashboard.[3]
Tools, writing & talks
Tools
Start here
OpenClaw Managed Agents
github.comA self-hosted API service around OpenClaw with isolated sessions and durable events.
Read the agent/session/event example to understand the API shape. Trying it requires Docker, a supported model-provider key and a local or cloud host you control.
Creator and repository owner[1][2]Original source- Hermes Labyrinth
github.comA read-only journey inspector and report exporter for Hermes Agent.