Beta — CelerFlow currently supports OpenClaw agents. MCP Gateway support for other frameworks is coming soon.
What is CelerFlow?
Other tools show you what agents do. CelerFlow lets you control what they CAN do. CelerFlow is a governance platform for AI agents. One command to connect. Full visibility into every tool call. Service-level permission controls. A kill switch when things go wrong. Human-in-the-loop approval for dangerous operations.Tracing
Every tool call is recorded and classified — outbound writes are highlighted, local reads are hidden by default. Traces are grouped by task for easy browsing.
Service permissions
Control access at the service level (Gmail, Notion, GitHub…) with separate read/write toggles. Write operations can require human approval.
Kill switch
Terminate any agent instantly from the dashboard. All tool calls are blocked within seconds via real-time push.
Risk alerts
Automatic detection of anomalous behavior — token spikes, high block rates, latency anomalies, agents going offline.
How it works
Connect your agent
Run
celerflow connect --code <setup-code>. The CLI discovers your OpenClaw agents, registers them, and writes the plugin config automatically.Traces flow automatically
Every tool call is recorded with category (outbound write, local read, etc.), service name, parameters, and result. Traces are grouped by task/session.
Set service permissions
From the dashboard, configure per-service read/write permissions. Flag dangerous write operations for human approval.
Why not just observability?
| Observability tools | CelerFlow | |
|---|---|---|
| See what agents do | ✅ | ✅ |
| Control what agents can access | ❌ | ✅ Service-level permissions |
| Stop an agent instantly | ❌ | ✅ Kill switch |
| Approve dangerous operations | ❌ | ✅ Human-in-the-loop |
| Detect anomalous behavior | ⚠️ Manual alerts | ✅ Automatic risk alerts |
| Agent health scoring | ❌ | ✅ 0–100 score with trends |
Get started
Quickstart
Connect your first agent and see traces in under 5 minutes.