OpenClaw + Faramesh Integration
Faramesh provides a first-class plugin for OpenClaw that intercepts every tool call and routes it through the Faramesh policy engine for Allow, Ask (human approval), or Deny decisions, before the tool executes.
What Faramesh adds to OpenClaw
Capability | OpenClaw alone | With Faramesh |
|---|---|---|
Tool execution | Yes | Governed tool execution |
Policy enforcement | Basic | Full category-based and rule-based policies |
Human approval | No | Slack, email, and dashboard approval workflows |
Audit trail | Partial | Complete: every decision, every tool call |
Multi-runtime visibility | No | Fleet management: see all runtimes under one org |
Per-agent policies | No | Each agent can have its own Allow/Ask/Deny settings |
CLI control | No |
|
Architecture
The Faramesh plugin registers a
before_tool_callhook at priority 1000 (runs first).Every tool call (bash, file read/write, browser, network, etc.) is sent to the Faramesh server.
The server evaluates the action against the active policy (simple category-based or advanced YAML/JSON).
The result is returned: allow (tool runs), require_approval (tool blocked until human approves), or deny (tool blocked permanently).
If approval is required, notifications are sent via Slack and/or email. The user can approve/deny from the dashboard, Slack, or CLI.
Fail-closed by default
If the Faramesh server is unreachable, the plugin blocks all tool calls by default (fail-closed). This can be changed to fail-open in the plugin configuration.
Quick start
See the Install guide for full instructions.
