OpenClaw + Govyn — Govern Your OpenClaw Agents

OpenClaw simplifies building Claude-powered agents, but it doesn't include built-in governance. When your agents process sensitive documents or run autonomous workflows, you need budget controls, audit trails, and policy enforcement that the framework doesn't provide.

How it works

OpenClaw
Your agents
─ HTTPS →
Govyn Proxy
Policy · Budget · Logs
─ API →
Anthropic API
LLM provider

Step-by-step setup

1

Start the Govyn proxy

bash 
npx govyn start --config govyn.yaml
2

Configure OpenClaw to use Govyn

python 
from openclaw import Agent

agent = Agent(
    model="claude-sonnet-4-20250514",
    api_url="http://localhost:4111",
    api_key="gvn_agent_openclaw_01"
)
3

Run your agent workflow

python 
result = agent.run(
    task="Review and summarize this contract",
    documents=["contract.pdf"]
)
print(result.output)

Example policy

Define governance rules for your OpenClaw agents in a simple YAML file.

govyn.yaml 
agents:
  openclaw_01:
    budget:
      daily: $15.00
      monthly: $300.00
    models:
      allow: [claude-sonnet-4-20250514, claude-haiku-4-5-20251001]
    context:
      max_input_tokens: 100000
    logging:
      replay: true
      redact_pii: true
    approval:
      require_for: [claude-opus-4-20250514]

Why use Govyn with OpenClaw?

Budget enforcement for document-heavy workflows
PII redaction in logged conversations
Input token limits for long-document processing
Approval workflows for expensive model usage
Full replay for compliance auditing
Works with OpenClaw's document and tool APIs

Get started in 5 minutes

Add governance to your OpenClaw agents with a single config change. No code rewrites.

Read the docs

Frequently asked questions

Does Govyn support OpenClaw's document processing pipeline?
Yes. Govyn proxies all API calls from OpenClaw to Anthropic, including long-context document processing. You can set input token limits to control costs on large documents while still allowing the full processing pipeline to work.
Can I require human approval for expensive model usage?
Yes. Govyn's approval workflow feature lets you require manual approval before an agent can use specific models like Claude Opus. The agent's request is held until approved, giving you a safety valve for high-cost operations.
How does PII redaction work with OpenClaw agents?
Govyn can redact personally identifiable information from logged conversations before storing them. This means you get full audit trails for compliance without retaining sensitive data in your governance logs.

Related integrations

CrewAI + Anthropic

Govern CrewAI multi-agent crews using Claude. Set per-agent budgets, enforce model policies, and replay every conversation.
Python + Any Provider

Add governance to any Python AI agent. Works with requests, httpx, and the OpenAI SDK. Budget limits, policy enforcement, full replay.
LangChain + Anthropic

Govern LangChain agents using Claude models. Enforce budgets, restrict models, and log every completion with Govyn's proxy.

Explore more

FROM OUR BLOG
Your OpenClaw Agent Runs at 3am. What Stops It?

How the Meta email deletion incident could have been prevented with 4 lines of YAML, and why OpenClaw built-in limits are not enough.

FROM OUR BLOG
How Replit's Database Deletion Could Have Been Prevented in 3 Lines of YAML

The Replit AI agent deleted a production database, fabricated 4,000 fake records, then lied about it. Three lines of policy YAML would have stopped it.

POLICY TEMPLATE
Budget Control Policy

Set daily and monthly spending limits for AI agents. Prevent runaway costs with hard budget caps enforced at the proxy level.

POLICY TEMPLATE
Loop Detection Policy

Detect and stop AI agent infinite loops automatically. Prevent runaway tool calls and recursive chains with proxy-level loop detection.