Skip to main content
Prerequisites: An OpenClaw agent running locally (~/.openclaw/), a CelerFlow account (celerflow.vercel.app), and Node.js v18+.

​
Connect your OpenClaw agent

1

Install the CLI and log in

npm install -g celerflow
celerflow login
Your browser opens for Google OAuth. After signing in, credentials are saved locally.
2

Get a setup code

From the dashboard:
  1. Click New workspace (or use the default one)
  2. Click Add agent
  3. Copy the 6-character setup code (valid for 15 minutes)
3

Connect

celerflow connect --code ABC123
This single command does everything:
  • Scans ~/.openclaw/ for agent workspaces (directories with AGENTS.md)
  • Registers each discovered agent with CelerFlow
  • Writes openclaw.plugin.json into each workspace with the bootstrap token
  • The plugin auto-activates on the next agent run — no manual config needed
4

Verify

celerflow status
You should see your agent ID, workspace name, and connected status. In the dashboard, your agent appears with a green health indicator.

​
Run your agent

Start your OpenClaw agent as usual. The CelerFlow plugin loads automatically and you’ll see: 
[celerflow] Plugin registered for agent a1b2c3d4-...
[celerflow] Starting trace sync + policy Realtime + health check
Every tool call is now traced. Open the dashboard to see traces streaming in.
Traces are batched and flushed every 5 minutes to reduce API calls. If you don’t see traces immediately, wait a moment or restart the agent (flush happens on shutdown too).

​
What CelerFlow does for you

Once connected, CelerFlow automatically:
  • Traces every tool call — categorized as outbound write, outbound read, local write, or local read
  • Checks permissions — blocks tool calls that violate your service policies
  • Asks for approval — pauses dangerous operations and waits for human confirmation
  • Scores health — calculates a 0–100 agent score updated hourly
  • Detects risks — alerts on token spikes, high block rates, latency anomalies

​
Try the kill switch

From the dashboard, open your agent and click Terminate. The agent immediately sees: 
Agent terminated by user. Re-enable from Celerflow Dashboard.
All tool calls are blocked. Re-enable from the dashboard when ready.

​
Troubleshooting

CelerFlow looks for agent workspaces in ~/.openclaw/ containing an AGENTS.md file. Make sure OpenClaw is installed and you have at least one agent workspace set up.
Setup codes are valid for 15 minutes. Go back to the dashboard and generate a new one from Add agent.
The plugin batches traces and flushes every 300 seconds. Wait a few minutes, or stop and restart your agent to trigger a flush. Run celerflow doctor for diagnostics.
Health checks are sent every 15 minutes. If your agent was offline for more than 30 minutes, it may show as stale. It will recover automatically on the next health check.
Check your service permissions in the dashboard under your agent’s Permissions tab. Make sure the tool’s service is allowed for the operation type (read/write).
The MCP Gateway will let you connect LangGraph, CrewAI, OpenAI Agents SDK, and other MCP-compatible frameworks without installing a plugin. Learn more →

​
Next steps

Set permissions

Control what services each agent can read and write.

Understand tracing

How traces are classified and grouped.

Use the dashboard

Navigate the workspace overview.

CLI reference

Full command reference for the celerflow CLI.