# The GOVENANT Standard

> GOVENANT — govern + covenant: governed autonomy, bound as a covenant. The open standard
> (CC BY) for provable, governed AI agents: actions prevented by construction, outcomes
> verified in the substrate, coverage diffed daily so silence is detectable, autonomy earned
> on evidence.
> GOVENANT implements the OCMAS formalism (Ferber, Gutknecht & Michel) on LLM substrates.
> Built on the Organization-Centered MAS / Agent–Group–Role (AGR) lineage — credited at
> https://govenantstandard.org/acknowledgments/

## Run the Audit — free, hosted, read-only (the GOVENANT Audit)
> Score your own AI agents against the standard in minutes. Connect the hosted MCP server to the AI
> tool you already use (bring your own agent), sign in with GitHub or Google, and your agent runs a
> read-only, aggregate-only sweep — detecting performed autonomy (motion without verified delivery)
> and grading you on the GOVENANT ladder: Logged → Gated → Delivered → Earned. Never reads prompts,
> agent I/O, customer data, or secrets. Free forever under the covenant.
> MCP server URL: https://mcp.govenantstandard.org/mcp
- [Run the Audit (all tools + FAQ)](https://govenantstandard.org/audit/)
- [Audit your Claude agents](https://govenantstandard.org/audit/claude/): add the GOVENANT Audit as a custom connector in the Claude desktop app.
- [Audit your ChatGPT agents](https://govenantstandard.org/audit/chatgpt/): add it in ChatGPT developer mode (Apps & Connectors → Advanced → OAuth).
- [Audit your Claude Code agents](https://govenantstandard.org/audit/claude-code/): one command — `claude mcp add --transport http govenant`.
- [Audit your Cursor agents](https://govenantstandard.org/audit/cursor/): add the server to `~/.cursor/mcp.json`.
- [Audit your VS Code (Copilot) agents](https://govenantstandard.org/audit/vscode/): add it to `.vscode/mcp.json` (the `servers` key, `type: http`).
- [Audit your Windsurf agents](https://govenantstandard.org/audit/windsurf/): add it to `~/.codeium/windsurf/mcp_config.json` (the `serverUrl` key).
- [Audit your Codex CLI agents](https://govenantstandard.org/audit/codex/): `codex mcp add govenant --url …` then `codex mcp login`.
- [Audit your Cline agents](https://govenantstandard.org/audit/cline/): use the `mcp-remote` bridge in `cline_mcp_settings.json`.

## Integrations — govern any agent stack
> GOVENANT governs substrates, not frameworks: any stack whose actions can be recorded and
> verified can be governed. One pattern page per framework / platform / vendor / protocol,
> each with the value proposition, the class-level integration pattern, and the public GitHub
> issue where the adapter or schema mapping is built in the open.
- [All integrations (45 stacks, 6 classes)](https://govenantstandard.org/integrations/)
- [Govern your LangChain / LangGraph agents](https://govenantstandard.org/integrations/langchain/): LangGraph's checkpointer and node/edge graph give natural interception points; callbacks already carry run IDs, token usage, and tool I/O — everything an action row needs.
- [Govern your OpenAI Agents SDK agents](https://govenantstandard.org/integrations/openai-agents-sdk/): The SDK's guardrails and handoffs map to ownership gates, and its tracing hooks carry tool-call inputs and outputs for the ledger.
- [Govern your Claude Agent SDK agents](https://govenantstandard.org/integrations/claude-agent-sdk/): PreToolUse and PostToolUse hooks are a ready-made chokepoint: check ownership before the call executes, write the action row after it returns.
- [Govern your AutoGen / AG2 agents](https://govenantstandard.org/integrations/autogen/): The conversable-agent message loop is a typed event bus waiting for named consumers; registered reply functions are the chokepoint where the gate and ledger live.
- [Govern your Microsoft Semantic Kernel agents](https://govenantstandard.org/integrations/semantic-kernel/): Kernel function-invocation filters intercept every skill call — the ownership gate and the ledger write fit in a single filter.
- [Govern your CrewAI agents](https://govenantstandard.org/integrations/crewai/): Crews, roles, and tasks map one-to-one onto GOVENANT roles and duties; step and task callbacks provide the ledger writes. A crew definition is a duty roster waiting to be made accountable.
- [Govern your LlamaIndex agents](https://govenantstandard.org/integrations/llamaindex/): Workflow steps and instrumentation spans give the interception points; tool outputs become verified outcome rows instead of trusted return values.
- [Govern your Haystack agents](https://govenantstandard.org/integrations/haystack/): Pipelines are explicit DAGs — wrap components at the pipeline boundary and every run is ledgered with its inputs, outputs, and cost.
- [Govern your PydanticAI agents](https://govenantstandard.org/integrations/pydantic-ai/): Typed, validated outputs are half of the deterministic validation gate already; result validators and tool wrappers supply the ownership gate and the ledger.
- [Govern your Mastra agents](https://govenantstandard.org/integrations/mastra/): The workflow engine's steps and retries map to duties and breakers, and its storage layer can host the substrate tables directly.
- [Govern your Vercel AI SDK agents](https://govenantstandard.org/integrations/vercel-ai-sdk/): Tool-call middleware and onStepFinish callbacks are the chokepoint; the ledger ships to any SQL store your app already uses.
- [Govern your smolagents (Hugging Face) agents](https://govenantstandard.org/integrations/smolagents/): Step callbacks and the tool abstraction give the hooks. Small agents deserve real records too — a governed smolagent outranks an ungoverned fleet.
- [Govern your Google Agent Development Kit (ADK) agents](https://govenantstandard.org/integrations/google-adk/): Before- and after-tool callbacks plus session state give the gate and ledger points; on Vertex, the substrate lives in Cloud SQL and the sweep is read-only SQL.
- [Govern your Claude Code agents](https://govenantstandard.org/integrations/claude-code/): Hooks make ownership gates enforceable per tool call, and the hosted GOVENANT Audit connects to Claude Code today as an MCP server — this is the shortest path in the whole catalog.
- [Govern your Cursor agents](https://govenantstandard.org/integrations/cursor/): Cursor speaks MCP, so the hosted GOVENANT Audit connects today; rules plus a governed substrate turn agent-written code into outcome-keyed, verified delivery.
- [Govern your GitHub Copilot coding agent agents](https://govenantstandard.org/integrations/github-copilot/): Copilot's pull requests are verifiable terminal outcomes. Govern the repo, not the model: branch protections act as ownership gates, and merged-plus-green is the outcome row.
- [Govern your Devin (Cognition) agents](https://govenantstandard.org/integrations/devin/): Devin's sessions end in pull requests — auditable terminal outcomes. Wrap its task dispatch in a duty roster so every assigned task has an owner and a verified completion.
- [Govern your OpenAI Codex agents](https://govenantstandard.org/integrations/openai-codex/): Codex tasks end in diffs and pull requests, so outcome-keyed completion is native; the Codex CLI also connects to the hosted GOVENANT Audit as an MCP client today.
- [Govern your Windsurf agents](https://govenantstandard.org/integrations/windsurf/): Windsurf's Cascade speaks MCP, so the hosted GOVENANT Audit connects today; governed substrate tables turn agentic flows into verified, ledgered delivery.
- [Govern your Priiism agents](https://govenantstandard.org/integrations/priiism/): Priiism operates as a governed tenant of the standard's reference implementation — the live existence proof for this entire class. Its definition of done lives in the database, not in the model's mouth.
- [Govern your Salesforce Agentforce agents](https://govenantstandard.org/integrations/agentforce/): Agentforce actions and flows already log to Salesforce objects; a schema mapping over those objects lets the sweep read what the platform already records — the org's own data becomes the substrate of record.
- [Govern your Microsoft Copilot Studio agents](https://govenantstandard.org/integrations/copilot-studio/): Topics and actions log to Dataverse; map those tables once and the battery runs read-only over what Power Platform already captures.
- [Govern your ServiceNow AI Agents agents](https://govenantstandard.org/integrations/servicenow/): The platform's workflow audit trail is substrate-shaped out of the box; map the tables, run the sweep, and put conformance verdicts next to the dashboards operators already trust.
- [Govern your Amazon Bedrock Agents agents](https://govenantstandard.org/integrations/bedrock-agents/): Trace events plus CloudWatch and CloudTrail records map to the action ledger; action groups are levers, with IAM as the ownership gate's enforcement arm.
- [Govern your Google Vertex AI Agent Builder agents](https://govenantstandard.org/integrations/vertex-agent-builder/): Agent traces and BigQuery logging make the schema mapping a SQL view; the sweep runs as read-only BigQuery, and verdicts land where your analysts already work.
- [Govern your Zapier Agents agents](https://govenantstandard.org/integrations/zapier-agents/): Zap runs and task history are an action ledger already; a mapping plus a duty roster over scheduled Zaps yields the coverage tests — including the misses.
- [Govern your n8n agents](https://govenantstandard.org/integrations/n8n/): The execution log in Postgres is directly sweepable: workflows become duties, executions become duty-runs, and error workflows become the independent watchdog.
- [Govern your Make agents](https://govenantstandard.org/integrations/make/): Scenario execution logs map to duty-runs, and incomplete executions become the honest miss log — the number most automation dashboards quietly omit.
- [Govern your Relevance AI agents](https://govenantstandard.org/integrations/relevance-ai/): Agent runs and tool logs map to the ledger; the multi-agent workforce framing is a duty roster asking to be made accountable, role by role.
- [Govern your Dust agents](https://govenantstandard.org/integrations/dust/): Assistant run logs map to action rows, and spaces with access controls align naturally with ownership gates.
- [Govern your Lindy agents](https://govenantstandard.org/integrations/lindy/): Each Lindy is a role with a charter; make the charter map to duties, key completion to verified outcomes, and the sweep tells you which Lindies actually deliver.
- [Govern your Gumloop agents](https://govenantstandard.org/integrations/gumloop/): Flow runs map to duty-runs and the node I/O record is the action ledger; governance is a mapping away, not a rebuild.
- [Govern your Vapi agents](https://govenantstandard.org/integrations/vapi/): Every call is a duty-run with a transcript, an end reason, and an outcome. Map call records to outcome rows and the coverage math runs on real conversations — misses included.
- [Govern your Retell AI agents](https://govenantstandard.org/integrations/retell/): Call analytics and post-call webhooks feed the outcome rows; missed and failed calls become the honest miss log that earns customer trust.
- [Govern your LiveKit Agents agents](https://govenantstandard.org/integrations/livekit-agents/): Session events and egress give the record; the standard upgrades "the call happened" to "the caller's need was verifiably handled."
- [Govern your ElevenLabs Agents agents](https://govenantstandard.org/integrations/elevenlabs-agents/): Conversation history and evaluation criteria map to outcome rows and validation gates; every conversation gets graded against the record, not the vibe.
- [Govern your Sierra agents](https://govenantstandard.org/integrations/sierra/): Outcome-based pricing already prices exactly what GOVENANT measures — verified resolution. A conformance record makes that claim inspectable by every buyer.
- [Govern your Decagon agents](https://govenantstandard.org/integrations/decagon/): Resolution metrics become sweepable outcome rows, and the independent watchdog writes down what a dashboard would rather smooth over.
- [Govern your Intercom Fin agents](https://govenantstandard.org/integrations/intercom-fin/): Fin's resolutions feed outcome rows; the delta between "answered" and "resolved, verified against the record" is precisely the performed-autonomy gap the audit measures.
- [Govern your UiPath agents](https://govenantstandard.org/integrations/uipath/): Orchestrator is substrate-shaped: queues are duty rosters, job results are duty-runs, exceptions are the miss log. Agentic automation inherits a governance record RPA teams already half-keep.
- [Govern your Automation Anywhere agents](https://govenantstandard.org/integrations/automation-anywhere/): Control Room audit logs map to the action ledger, and bot schedules are duties with outcome-keyed completion.
- [Govern your Temporal agents](https://govenantstandard.org/integrations/temporal/): Temporal's event history IS an append-only action ledger — likely the cheapest path to Tier-1 conformance in this catalog. Workflows are duties; activities are levers.
- [Govern your Inngest agents](https://govenantstandard.org/integrations/inngest/): Function runs, steps, and flow control give the record; cron functions are duties and the failure path is the miss log, kept honestly.
- [Govern your Model Context Protocol (MCP) agents](https://govenantstandard.org/integrations/mcp/): A governing MCP gateway ledgers every tool call and enforces ownership gates for any MCP-speaking agent, regardless of framework — govern the pipes once, govern every client.
- [Govern your Agent2Agent Protocol (A2A) agents](https://govenantstandard.org/integrations/a2a/): A2A tasks and artifacts are typed events with named consumers — the standard's event-bus requirement expressed as an interop protocol. Ledger at the task boundary and inter-agent work becomes provable.

## Performed autonomy — the diagnosis
> The coined term for the failure mode: an AI agent system that produces the appearance of
> governed work — activity in the logs, "done" in the record — with no verifiable outcome
> underneath. Definitional page + the six plain-English failure patterns (each mapped to the
> standard's official anti-pattern catalog).
- [Performed autonomy — definition, origin, press kit](https://govenantstandard.org/performed-autonomy/)
- [The self-reported outcome](https://govenantstandard.org/performed-autonomy/self-reported-outcome/): the agent writes "task complete" and nothing independent checks.
- [The prompt-shaped guardrail](https://govenantstandard.org/performed-autonomy/prompt-shaped-guardrail/): a rule in a system prompt is a preference, not a wall.
- [The author-reviewer](https://govenantstandard.org/performed-autonomy/author-reviewer/): the same model writes the work and reviews the work.
- [The silent duty](https://govenantstandard.org/performed-autonomy/silent-duty/): a job stopped firing weeks ago and nothing noticed.
- [The ungraded prediction](https://govenantstandard.org/performed-autonomy/ungraded-prediction/): decisions logged, results never measured.
- [The toggle](https://govenantstandard.org/performed-autonomy/the-toggle/): autonomy granted by a setting, not earned on evidence.
- [The three audits — we failed our own audit twice, and published both](https://govenantstandard.org/three-audits/)
- [The Reality Check — free probe, level 1–4](https://govenantstandard.org/reality-check/)
- [The Expert Audit — done with you, from $9,500](https://govenantstandard.org/expert-audit/)
- [Press kit — definitions, diagrams, citation](https://govenantstandard.org/press/)
- [Partners — implementation & referral, get started](https://govenantstandard.org/partners/)

## The Standard (normative)
- [Part 0: The Standard](https://govenantstandard.org/standard/00-standard/)
- [Part 1: The Organization Model](https://govenantstandard.org/standard/01-org-model/)
- [Part 2: The Role Contract](https://govenantstandard.org/standard/02-role-contract/)
- [Part 3: The Duty Roster ⭐](https://govenantstandard.org/standard/03-duty-roster/)
- [Part 4: The Substrate](https://govenantstandard.org/standard/04-substrate/)
- [Part 5: The Life of a Governed Action](https://govenantstandard.org/standard/05-governed-action/)
- [Part 6: The Accountability Loop](https://govenantstandard.org/standard/06-accountability/)
- [Part 7: Earned Autonomy and Resilience](https://govenantstandard.org/standard/07-autonomy-and-resilience/)
- [Part 8: The Constitution Plane](https://govenantstandard.org/standard/08-constitution/)
- [Part 9: The Compliance Audit Instrument](https://govenantstandard.org/standard/09-audit-instrument/)
- [Part 10: The Anti-Patterns](https://govenantstandard.org/standard/10-anti-patterns/)
- [Part 11: The UI Surface (the Ant Farm)](https://govenantstandard.org/standard/11-ui-surfaces/)
- [Part 12: The Porting Guide](https://govenantstandard.org/standard/12-porting-guide/)
- [Part 15: Developer Integration](https://govenantstandard.org/standard/15-developer-integration/)
- [Part 16: The Repo Scanner and Continuous Certification](https://govenantstandard.org/standard/16-repo-scanner/)
- [Anti-Pattern Reconciliation — campaign vocabulary ↔ the standard](https://govenantstandard.org/standard/anti-pattern-reconciliation/)
- [NAMING — The Mark, the Architecture, the Rules](https://govenantstandard.org/standard/naming/)
- [The Covenant](https://govenantstandard.org/standard/the-covenant/)

## Plain-English explainer
- [1. The Big Idea](https://govenantstandard.org/learn/01-the-big-idea/)
- [2. The Problem](https://govenantstandard.org/learn/02-the-problem/)
- [3. How It Works](https://govenantstandard.org/learn/03-how-it-works/)
- [4. The Idea Behind It](https://govenantstandard.org/learn/04-the-idea-behind-it/)
- [8. Word List](https://govenantstandard.org/learn/08-word-list/)

## Full text
- [llms-full.txt](https://govenantstandard.org/llms-full.txt): the complete standard and explainer in one file