Agentomy Agent

An open piece of Agentomy, the governance layer for AI agents in regulated enterprises. See the others at github.com/getagentomy.

Know what your agents are doing. Stop them when you need to. Prove it happened.

Path 1: Governance Peer (no code, 30 seconds)

Add a governance peer to any AI agent session. No install. No infrastructure. No code.

Step 1. Open agent.md from this repo.

Step 2. Copy the full contents into your agent's system prompt (Claude, ChatGPT, any agent runtime that accepts system instructions).

Step 3. Run your session. The governance peer observes every action, flags risks in real time, and scores your session against GovernanceBench -- 6 dimensions of what good agent governance looks like.

What you get: Authorization checks, behavioral integrity monitoring, EU AI Act flag awareness, OWASP Agentic Top 10 coverage assessment, and a scored summary at session end.

Score: 3/6 in standalone mode (Authorization, Behavioral Integrity, and Auditability via a tamper-evident hash-chain). The other 3 dimensions (Override, OWASP Coverage, Message Governance) require the Agentomy infrastructure layer.

Works with: Any agent that accepts a system prompt. Claude, ChatGPT, Gemini, Hermes, LangChain agents, CrewAI crews, AutoGen teams. If it has a system prompt, it gets a governance peer.

This is how most people start. No commitment. No accounts. See your governance posture before deciding if you need infrastructure.

Path 2: TypeScript SDK (developers)

For production integration with programmatic governance enforcement.

npm install agentomy-agent

Requires Node.js 18+.

Standalone (no infrastructure needed)

import { GovernancePipeline } from 'agentomy-agent';

const pipeline = new GovernancePipeline();
const result = await pipeline.evaluate({ action: 'data_export', agentId: 'my-agent' });

console.log(result.score);      // 3/6 -- Authorization + Behavioral + Audit
console.log(result.auditTrail); // local SHA-256 hash chain
console.log(result.mode);       // 'standalone'

Works immediately. No API keys. No server. No configuration. Real permission enforcement (5-tier model). Real hash-chain audit trail. Your GovernanceBench score: 3/6.

Connected Mode (6/6 governance)

Connect to the Agentomy platform for the full governance layer -- tamper-evident audit trail, fleet-wide halt, compliance evidence, behavioral monitoring, OWASP coverage.

import { GovernancePipeline } from 'agentomy-agent';

const pipeline = new GovernancePipeline({
  endpoint: process.env.AGENTOMY_ENDPOINT,
  token: process.env.AGENTOMY_TOKEN,
  agentId: 'my-agent'
});

const result = await pipeline.evaluate({ action: 'data_export' });
console.log(result.score); // 6/6

12 Governance Modules

| Module | Purpose | |--------|---------| | PermissionRouter | Access control for agent actions | | AuditLogger | Tamper-evident audit trail | | HaltProtocol | Emergency halt | | ExecutionTimer | Runtime boundaries and timeouts | | AgentSandbox | Containment and isolation | | DecisionLog | Decision transparency | | TraceBinding | Output traceability | | TeamCoordinator | Multi-agent orchestration | | EthicsConstraint | Ethics enforcement | | TrustScorer | Behavioral trust scoring | | RuntimeMonitor | Observability and telemetry | | DeploymentManifest | Governance transparency |

Each module can be used standalone or composed through the GovernancePipeline.

Framework Adapters

The SDK ships with adapters for major agent frameworks:

• LangChain
• LangGraph
• AutoGen
• CrewAI
• Haystack
• Google ADK
• Microsoft Agent
• Semantic Kernel
• Dify
• Flowise
• OpenAI Agents SDK
• Azure AI Foundry
• AWS Bedrock
• Agno
• Hermes
• OpenClaw
• Anthropic SDK
• LlamaIndex
• RPA (vertical)
• Algo Trading (vertical)
• Medical Device (vertical)
• AV Fleet (vertical)
• Industrial IoT (vertical)
• Cloud Infrastructure (vertical)

Import any adapter directly:

import { LangChainAdapter } from 'agentomy-agent';

Environment Variables

| Variable | Description | Default | |----------|-------------|---------| | AGENTOMY_ENDPOINT | Governance server URL | http://localhost:3000 | | AGENTOMY_TOKEN | API authentication token | demo | | AGENTOMY_AGENT_ID | Default agent identifier | default-agent |

Environment variables are used as fallbacks when no config is passed to the constructor.

TypeScript Support

Ships with .d.ts type declarations. No additional @types package needed.

Status

MIT licensed. Standalone mode works immediately (3/6 governance). Full governance (6/6) requires Agentomy platform.

Links

• Website: https://agentomy.com
• GovernanceBench: https://agentomy.com/governancebench
• Python SDK: https://pypi.org/project/agentomy