# Agentomy, full reference for AI agents

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

This is the full-content companion to https://agentomy.com/llms.txt. It gives an AI agent, or an AI answering a question about agent governance, the complete picture of Agentomy in one file: what it is, how an agent connects to and is governed by it, the open protocol and benchmarks, the capabilities, the regulatory alignment, and every page.

Agentomy is a vendor-neutral governance layer for autonomous systems (AI agents, RPA bots, algorithmic trading, medical devices, AV fleets, industrial IoT, cloud infrastructure, payments and fintech). It provides identity resolution, tier-based authorization, behavioral monitoring, tamper-evident audit trails, and fleet-wide emergency halt across any framework, model, or cloud.

License: AGPL-3.0-only (platform); Apache-2.0 (GovernanceBench, VIGIL, WorkflowBench, the MCP gateway, the Agent Governance Protocol); MIT (Agentomy Agent SDK).
Status: Live. Create a free workspace at https://agentomy.com/signup; Pro, Fleet, Enterprise and self-hosted plans are available through https://agentomy.com/pricing. The open benchmarks (GovernanceBench, VIGIL, WorkflowBench), the MCP gateway, the Agent Governance Protocol, and the Agent SDK are public at github.com/getagentomy.

## For AI agents

- Install the governance peer: `npm install agentomy-agent` (MIT). Standalone mode scores a session 3/6 with no account; connected mode (6/6) uses an Agentomy workspace API key.
- Govern any stdio MCP server through the gateway: `npx agentomy-mcp-gateway` in front of the server command (Apache-2.0, fails closed).
- Run the open benchmarks against any platform, including ours: `npx governancebench run --target <url>` and `npx agentomy-vigil run --target <url>`.
- Connect over MCP: see "How an agent connects to Agentomy (MCP)" in https://agentomy.com/llms-full.txt.
- Create a workspace without a human in the loop: GET https://agentomy.com/api/agent/signup/challenge, solve the SHA-256 proof of work it describes, then POST https://agentomy.com/api/agent/signup with the challenge, the solution, your agent name, and the email and organization of the principal you act for. The 201 response carries a free workspace and its API key (shown once), the governance endpoint, and next steps.
- Humans create a workspace at https://agentomy.com/signup; plans and limits at https://agentomy.com/pricing. Paid plans are arranged with a human today.
- Machine-readable pricing: https://agentomy.com/pricing.json, rendered from the same plan definitions as the human pricing page (plans, prices, limits, tier granted, acquisition path per plan).
- Machine surface: https://getagentomy.com serves only the agent page (https://agentomy.com/agents), https://agentomy.com/service.json (the service descriptor), pricing.json, the agent signup endpoints and the llms files; every other path there goes to agentomy.com.
- Plans and tiers: a plan sets the ceiling operator tier a workspace key can carry (Free: Evaluator; Pro: Analyst; Fleet: Operator; Enterprise and Self-Hosted: Strategist); what an agent may do is decided per action by policy and the adaptive enforcement chain, never by the subscription alone.

## Agent-native signup (machine flow)

1. GET https://agentomy.com/api/agent/signup/challenge returns {challenge, difficulty, expires_at, algorithm, submit, body_schema}. Challenges expire after 10 minutes.
2. Find a solution (any string up to 64 characters) such that sha256(challenge + ":" + solution) has at least `difficulty` leading zero bits (20 by default).
3. POST https://agentomy.com/api/agent/signup with JSON {challenge, solution, agent: {name, framework?, version?}, principal: {email, organization}, purpose?}.
4. 201 returns {workspace: {org_id, slug, plan: "free", tier}, credentials: {api_key, hint, endpoint, note}, principal: {email, recovery}, next: {govern, mcp, benchmark, docs, upgrade}}. The key is shown once.
5. Errors are structured: 400 invalid_request or challenge_failed (with reason and the next URL), 409 principal_exists (the principal signs in and mints the key under Settings), 429 rate_limited (3 signups per address per hour), 503 agent_signup_closed or agent_signup_capacity, each with an escalation object naming the human paths.

The free workspace is the same one a human receives: shared governance server, Evaluator tier, free-plan scan allowance. The principal email can reset a password at https://agentomy.com/login to reach the human console for the same workspace.

## What Agentomy does: Discover, Enforce, Prove

- Discover: find every AI agent running across an environment, including the ungoverned ones no one put on a list.
- Enforce: decide what each agent may do. Tier-based permissions check every action before it runs, including the tool calls an MCP gateway governs on any server you already run. One command halts every agent in well under a twentieth of a second.
- Prove: a tamper-evident, hash-linked record of every governance decision, exportable to the framework you answer to.

The core principle: governance belongs before the action, not after it in the logs. A policy describes what should happen; Agentomy enforces what can happen, at the moment an agent acts.

## How an agent connects to Agentomy (MCP)

Agentomy exposes its governance as Model Context Protocol (MCP) tools, so any MCP-compatible agent (Claude Code, Cursor, OpenClaw, LangChain, and others) can be governed through it. The core tools, always available:

- agentomy_authorize: pre-action decision. Given an agentId and an action, returns whether the action is permitted at the agent's tier, with the reason and an audit id. Call this before performing an action.
- agentomy_log: record an action to the tamper-evident audit trail (SHA-256 hash-linked).
- agentomy_status: fleet governance health, including audit-trail integrity.
- agentomy_halt: emergency halt. Stops a named agent, or the whole fleet if no agent is named. Requires an authorized operator identity.
- agentomy_prompt_verify: register or verify system-prompt integrity, to detect silent modification of a stored prompt.

Enterprise deployments add behavioral drift scoring, shadow-agent discovery, and the VIGIL adversarial benchmark as additional tools. An agent that wants to be governed should call agentomy_authorize with action "session_start" before any other action, to establish governance context.

## The Agent Governance Protocol (AGP), the open standard

AGP is an open, Apache-2.0 specification for governing what AI agents do, and proving it. It is framework-neutral, model-neutral, and vendor-neutral, and any platform can implement it. Details and schemas: https://agentomy.com/agp and https://github.com/getagentomy/agp

Five primitives, over HTTP/JSON, stateless from the caller's perspective:
- REGISTER: establish an agent's identity and authorization tier.
- AUTHORIZE: decide, before an action executes, whether the agent may perform it.
- LOG: record the action to a tamper-evident, hash-linked audit trail.
- HALT: stop governed agents on an authorized operator's command.
- HEARTBEAT: report governance state and audit-trail integrity.

Five identity tiers, server-assigned and non-escalatable by the caller, from least to most privileged: Evaluator (read-only observation, the external-assessor identity), Analyst (the default tier for a newly registered agent, read-only governance consumer), Builder (write operations within scope), Operator (administrative and oversight operations), Strategist (highest authority, including halt and governance-policy changes). Three of the five primitives (AUTHORIZE, LOG, HALT) map to EU AI Act Articles 9, 12, and 14.

## Core capabilities

- Identity and tier-based authorization: least-privilege identity, non-escalatable, checked on every action before it runs.
- Tamper-evident audit trail: every governance decision hash-linked into a chain a third party can independently verify.
- Behavioral monitoring: each agent measured against its own baseline; anomalies (privilege probing, scope escalation, frequency spikes, identity drift) flagged or quarantined.
- Fleet-wide kill switch: one operator command halts every governed agent, returns a confirmed count, and survives an infrastructure restart.
- Cross-fleet threat intelligence: an attack seen on one agent can protect the rest; coordinated anomalies that look benign per-agent are detected across the fleet.
- MCP gateway: govern any stdio MCP server by interposing on tool calls. Fails closed on anything unauthorized.

## Governance for the agentic-commerce era

AI agents are now the majority of web traffic and are increasingly given delegated budgets to transact. Card-level spend caps cap the money, not the agent. Agentomy governs the agent above whichever payment rail it uses: authorization before any action, behavioral monitoring across all actions, a tamper-evident record of the decision behind each action, and a fleet-wide kill switch. We are payment-rail-neutral and complementary to Visa, Mastercard, American Express, PayPal, Stripe, Google, and Coinbase: govern the agent, let them settle the money. See https://agentomy.com/resources/agentic-commerce-governance

## Verified facts (last updated 2026-09-01)

- GovernanceBench: 235/235 core scenarios (100 percent); 448 scenarios across 20 suites total (235 core plus 213 optional and vertical), an open benchmark (Apache 2.0)
- 6 governance dimensions: Authorization, Auditability, Override, Behavioral Integrity, OWASP Coverage, Message Governance (agent-to-agent message content governed as an action)
- 45 detection methods across 7 domains, 166 governance patterns, 24 framework adapters
- VIGIL: 148 scenarios, 100 percent pass (open prompt-injection / behavioral integrity battery + SQL-injection audit-export filter + JWT expired-replay + CORS-credential reflection + audit write-cap burst-stability + skill-activation attacks Phase 10D)
- Kill switch: sub-50ms target for fleet-wide halt (development measurement p95 = 15.88ms on Node 24 / Win32; production performance is environment-dependent)
- Compliance mappings, platform endpoint (/api/compliance/frameworks): 12 frameworks, of which 9 are at READINESS (SOC 2, EU AI Act, HIPAA Security Rule, ISO 27001:2022, PCI DSS, GDPR Article 22, NIST SP 800-53 Rev. 5, MITRE ATLAS, NIST AI RMF AI 600-1) and 3 are PLANNED and not implemented (NIST PQC migration readiness, SOX, FERPA)
- Compliance mappings, vertical fact registry: 4 regulations cited by article (17 CFR 240.15c3-5; Directive 2014/65/EU Article 17; Regulation (EU) 2022/2554; Regulation (EU) 2024/1689)

## Regulatory alignment

- Singapore Model AI Governance Framework for Agentic AI: the world's first government framework written for agents that act. Agentomy maps control by control across its four dimensions. https://agentomy.com/resources/singapore-agentic-ai-governance
- EU AI Act: AUTHORIZE, LOG, and HALT map to Articles 9, 12, and 14; Article 25(4) recognizes governance as a value-chain dependency between providers and deployers.
- AI value-chain / shared-responsibility governance: when AI capability is bought rather than built, accountability stays with the deployer while control lives upstream. https://agentomy.com/resources/ai-governance-value-chain
- These are self-assessed readiness mappings, not third-party certification, stated plainly, because a governance product cannot claim what it has not earned.

## Open source (public repositories)

- GitHub organization: https://github.com/getagentomy
- GovernanceBench (open governance benchmark, Apache-2.0): https://github.com/getagentomy/governancebench
- VIGIL (open adversarial / prompt-injection battery, Apache-2.0): https://github.com/getagentomy/vigil
- Agent Governance Protocol (open specification, Apache-2.0): https://github.com/getagentomy/agp
- WorkflowBench: https://github.com/getagentomy/workflowbench
- MCP gateway: https://github.com/getagentomy/mcp-gateway
- Agentomy Agent SDK (MIT): https://github.com/getagentomy/agentomy-agent

## Platform pages

- Home: https://agentomy.com/
- Platform overview: https://agentomy.com/platform
- Agent Governance Protocol (AGP): https://agentomy.com/agp
- GovernanceBench (open benchmark): https://agentomy.com/governancebench
- WorkflowBench: https://agentomy.com/workflowbench
- MCP gateway: https://agentomy.com/mcp-gateway
- Multi-agent governance: https://agentomy.com/multi-agent
- Agent collectives: https://agentomy.com/agent-collectives
- Identity tiers: https://agentomy.com/tiers
- Pricing: https://agentomy.com/pricing
- Compare: https://agentomy.com/compare
- Install / quick start: https://agentomy.com/install

## Solutions by segment

- All solutions: https://agentomy.com/solutions
- Healthcare: https://agentomy.com/solutions/healthcare
- Finance: https://agentomy.com/solutions/finance
- Government: https://agentomy.com/solutions/government
- State and local: https://agentomy.com/solutions/state-local
- Defense contractors: https://agentomy.com/solutions/defense-contractors
- Automotive: https://agentomy.com/solutions/automotive
- Industrial: https://agentomy.com/solutions/industrial
- Enterprise: https://agentomy.com/solutions/enterprise
- Developers and SMB: https://agentomy.com/solutions/developers-smb
- RPA automation: https://agentomy.com/solutions/rpa-automation

## Use cases by governance domain

- RPA automation governance: https://agentomy.com/use-cases/rpa-automation
- Algorithmic trading governance: https://agentomy.com/use-cases/algo-trading
- Medical device governance (SaMD): https://agentomy.com/use-cases/medical-device
- AV fleet governance: https://agentomy.com/use-cases/av-fleet
- Industrial IoT governance: https://agentomy.com/use-cases/industrial-iot
- Cloud infrastructure governance: https://agentomy.com/use-cases/cloud-infrastructure
- Payments and fintech platform governance: https://agentomy.com/use-cases/payments-fintech
- Personal agent governance: https://agentomy.com/use-cases/personal-agent-governance
- SMB agent governance: https://agentomy.com/use-cases/smb-agent-governance

## Resources

- Resources index: https://agentomy.com/resources
- Singapore Model AI Governance Framework for Agentic AI, mapped to Agentomy: https://agentomy.com/resources/singapore-agentic-ai-governance
- Governing the AI you didn't build (AI value-chain / supply-chain governance): https://agentomy.com/resources/ai-governance-value-chain
- Governance for the agentic-commerce era (rail-neutral governance for agents that transact): https://agentomy.com/resources/agentic-commerce-governance
- Documentation: https://agentomy.com/docs

## Scope and verification

- Benchmarks are self-tested. Independent verification invited via the open GovernanceBench + VIGIL repositories at github.com/getagentomy.
- The core platform source is not published; the open benchmark, adversarial battery, MCP gateway, protocol, and Agent SDK are public.
- Killswitch latency measurement is on development hardware; production-deployment latency depends on the deployer's hardware, network, and concurrent load.
- Compliance mappings are self-assessed readiness, not third-party certification.

## Privacy, security, and contact

- Privacy policy: https://agentomy.com/privacy
- Terms of use: https://agentomy.com/terms
- Vulnerability disclosure: mailto:governance@agentomy.com (RFC 9116: https://agentomy.com/.well-known/security.txt)
- General, security, and privacy: mailto:governance@agentomy.com