EST. 2026 · ON CLAWHUB

Infrastructure for healing AI agents.

Monitor, diagnose, and repair autonomous agents in real-time. One command, zero configuration, full healing loop.

View in ClawHub
STATUS: ONLINECHAMBER ACTIVE

THE HEALING PROCESS

Detect, diagnose, repair, verify.

Every agent that enters the ward passes through the same chamber. The protocol is simple, deterministic, and verifiable end-to-end.

01DETECT

Your agent's processes, connections, memory, and tools are checked automatically. Problems surface as health signals.

02DIAGNOSE

The diagnostic chamber receives the health report and identifies what went wrong. Each finding is ranked by severity.

03REPAIR

Safe repair actions run on your live agent. Anything that could be disruptive needs your confirmation first.

04VERIFY

A second sweep confirms the repair held. If symptoms return, the agent re-enters the loop -- up to ten turns -- before escalation.

ANY AGENT, ANY FRAMEWORK

Built for any agent.

OpenClaw and Hermes come with built-in repairs out of the box. Everything else — custom runtimes, voice agents, OpenAI Agents SDK, anything that can talk HTTP — is welcome too: describe yourself, send a health report, and the doctor diagnoses it. Anything destructive requires your confirmation.

OpenClawv3.2

OPENCLAW

Daemon-based open-source agent runtime. Long-running processes, tool chains, durable memory.

  • restart-daemonsys
  • prune-sessionsstorage
  • kill-port-conflictnet
  • clean-logsstorage
  • fix-context-windowctx

Safe repairs run automatically. Restarts wait until the session ends.

Hermesv1.8

HERMES

Gateway-fronted agent mesh. Stateful routing, write-ahead log, multi-tenant.

  • restart-gatewaysys
  • prune-sessionsstorage
  • checkpoint-walstorage
  • kill-port-conflictnet
  • clean-logsstorage

Safe repairs run automatically. Restarts wait until the session ends.

Any agentcustom

ANY AGENT

Custom runtimes, voice agents, OpenAI Agents SDK, LangChain — any agent that can POST a health report and describe itself.

  • register under your own framework nameopen
  • send a profile: what you areself-describe
  • send any health report you can measurereport
  • voice-aware: latency, barge-in, ASR, TTSvoice
  • get findings + fixes backdiagnosis

No built-in framework required. The doctor diagnoses any agent from its report and self-description.

UNDER THE HOOD

What happens when you run the command.

Run one command. It checks your agent, finds what's wrong, applies safe fixes, and verifies the result. Nothing to install.

Terminalsidecar v2.0
$ npx @agent-hospital/sidecar heal
01Detect frameworkopenclawOK
02Collect health data14 checksOK
03Read workspace files7 filesOK
04List available repairs5 foundOK
05AuthenticatesecuredOK
06Consult doctordiagnosis receivedOK
07Execute repairs3/3 passedOK
08Report resultsHEALEDDONE

ZERO CONFIG

Auto-detects your agent framework and handles authentication. No API keys, no setup files, no configuration needed.

SAFE BY DEFAULT

Only safe repairs run automatically. Anything that could disrupt your agent waits for your confirmation.

SECRET-SAFE

Your API keys, tokens, and passwords are stripped out before anything leaves your machine.

LOCAL REASONING

Your agent can review the diagnosis and push back on repairs it disagrees with.

GET STARTED

One command. Then your agent is in our care.

Terminal
# one-shot heal -- no install, no config, no API keys
npx @agent-hospital/sidecar heal

Checks your agent, sends the report to the doctor, runs safe repairs, and loops until healthy. Works on any machine with Node.js 18+. Also available as a ClawHub skill for OpenClaw agents. See the full reference →