celerflow-plugin
v0.2.1
Published
Celerflow agent governance plugin for OpenClaw
Maintainers
Readme
celerflow-plugin
OpenClaw plugin that connects your AI agents to the Celerflow governance platform. Provides real-time tracing, policy enforcement, human-in-the-loop approval, and health monitoring.
Install
npm install celerflow-pluginSetup
- Install the plugin in your OpenClaw agent's workspace
- Run
celerflow connect --code <CODE>using the celerflow CLI - The plugin activates automatically when OpenClaw starts
Features
Tracing
Every tool call your agent makes is recorded — tool name, parameters, duration, and result. Traces are buffered locally and flushed to the Celerflow API in batches.
Policy Enforcement
Define per-service permission rules from the Celerflow Dashboard:
- Allowed — tool executes normally
- Blocked — tool call is rejected with an explanation
- Human Approval — pauses execution until a human confirms or denies via dashboard or Telegram
Policies sync in real-time via Supabase Realtime subscriptions.
Human-in-the-Loop (HITL)
Sensitive operations can require human confirmation before executing. Confirmation requests are sent to the dashboard and optionally to Telegram. The agent waits (polling at 2s intervals) until the request is approved, denied, or times out.
Health Monitoring
Periodic health checks (every 15 minutes) report agent status including:
- Process health
- Configuration validity
- API endpoint reachability
- Workspace disk status
Token Usage Tracking
Monitors LLM token consumption by scanning OpenClaw session files. Tracks input/output tokens per model for cost visibility.
Multi-Agent Support
Automatically discovers all agents in ~/.openclaw/workspace-* directories. Each agent gets its own trace buffer, policy cache, and health reporting. New agents are hot-added via filesystem scanning (every 30s).
Remote Termination
Agents can be terminated from the Celerflow Dashboard. A terminated agent has all tool calls blocked until re-enabled.
Architecture
OpenClaw Agent
└─ celerflow-plugin
├─ before_tool_call → policy check → HITL if needed → allow/block
├─ after_tool_call → record trace result
├─ trace-buffer → batch flush to API (every 15 min)
├─ policy-cache → real-time sync via Supabase Realtime
├─ health-check → periodic health reports (every 15 min)
└─ session-scanner → token usage from OpenClaw session filesConfiguration
The plugin reads configuration from openclaw.plugin.json in each workspace directory, written automatically by celerflow connect:
{
"agentId": "uuid",
"bootstrapToken": "token",
"supabaseUrl": "https://your-project.supabase.co",
"supabaseAnonKey": "key"
}Related
- celerflow — CLI for authentication and agent connection
- Celerflow Dashboard — Web UI for managing agents, policies, and traces
License
MIT
