Agent class · Instrumented boundary

Govern your multi-agent systems.

Wrap the loop. Ledger every act. Prove delivery.

LangGraph, CrewAI, AutoGen, the OpenAI and Claude Agent SDKs — code-first frameworks give you the hooks; GOVENANT rides them to make every tool call answer for itself. The framework keeps running your agents; the adapter makes their work provable.

Agent-agnostic by design

GOVENANT does not govern “Claude agents” or “OpenAI agents.” It governs the evidence trail of autonomous work. Keep the runner (the tool you audit from) separate from the system under audit (your agents):

What runs the audit

ClaudeChatGPTClaude CodeCursorCodexVS CodeWindsurfCline

An AI client you already use reads the open instrument and drives the read-only sweep. This is the runner — not the thing being judged.

What gets audited — incl. multi-agent frameworks

LangGraphCrewAIAutoGen / AG2OpenAI Agents SDKClaude Agent SDKPydanticAILlamaIndex

Any agent system with an observable record of actions and outcomes — whatever built it. The agent never has to "support GOVENANT."

Ground truth: the database + actions + outcomes + duties + gates — the substrate, never the logs’ self-report.

Why govern multi-agent frameworks

Your agents already act — now they answer

Every tool call becomes an append-only action row: input, output, cost, verified outcome. When a customer, an auditor, or your own CEO asks what the agents did, the answer is a query, not a vibe.

Governance without changing your stack

The adapter rides the framework’s existing callback and middleware points — no rewrite, no new runtime, no lock-in. The standard is open under CC BY 4.0 and the method is free forever under the covenant.

Catch performed autonomy before your customers do

Agents that look busy but don’t verifiably ship are this generation’s failure mode. The sweep measures verified outcomes against the record — never the logs’ self-report.

Where this bites — a real scenario

A data team ships a LangGraph system: a planner, three specialists, and a critic, handing off through a shared state graph in production.

The challenge

When an output is wrong, no one can say which node did what, whether the critic actually ran, or whether a tool call that “succeeded” changed anything downstream. The graph is a black box the moment it leaves a notebook.

What the record reveals

Riding the checkpointer and callbacks, every node’s tool calls become append-only action rows with verified outcomes. The record shows the critic silently short-circuited on 12% of runs and one specialist’s “done” never wrote to the warehouse — invisible before, a query now.

How it works

The substrate: The framework’s own callbacks and middleware: run IDs, tool inputs/outputs, and cost — everything an append-only action row needs.

The path — Instrumented boundary: Add a thin hook at the action boundary — no access to prompts, reasoning, or models — so every action and its outcome is recorded.

  1. Add the adapter (or follow the documented recipe) — it hooks the framework’s existing callback or middleware points.
  2. Register every tool your agents can touch as a lever with a named owner; unowned calls stop at the gate.
  3. Route actions through the append-only ledger: input, output, cost, and the verified outcome — not the trusted return value.
  4. Define duties — the recurring work your agents owe — and key completion to outcomes, not activity.
  5. Connect the free GOVENANT Audit and run a trust sweep; hand the report to whoever needs to trust the system.

The full requirements live in the open standard (CC BY 4.0) — the substrate shapes, the acceptance tests, and the conformance ladder your record is measured against.

What you can claim

From day one of a real integration you can wear the Built-on GOVENANT badge under its four rules — real integration, a disclosure line, a link to the standard, no certification implication. From there the ladder is public: Logged → Gated → Delivered → Earned. Levels are self-assessed against the open standard and published to the registry with their probe logs, open to anyone’s challenge — no authority hands out stamps, which is exactly why a published record means something.

Stacks that build multi-agent frameworks

Per-stack integration patterns for common ways to build multi-agent frameworks — each with its own natural hook into the substrate:

Building it a different way? Browse all integrations — or run the free audit from any MCP-capable tool.

Governing multi-agent frameworks — FAQ

Do I have to replace my framework?
No. GOVENANT governs the substrate — the durable record of what agents did and whether it verifiably delivered — not the framework. Your framework keeps running your agents; the adapter makes their work provable.
What does the integration cost?
The standard, the pattern, and the adapters are open source, and the audit is free — the method is free forever under the covenant. What’s paid is optional attestation: the witnessed, signed, hosted record of your results and the live badge.
Do my agents have to “support” GOVENANT?
No. GOVENANT governs the substrate — the record of what agents did and whether it verifiably delivered — not the agent. If the system of record around your agents can be read, or you can instrument the action boundary, they can be measured, whatever framework, model, or vendor built them.
What is “performed autonomy”?
Agents that look busy — fluent plans, satisfied logs — but don’t verifiably ship. Motion is not delivery. GOVENANT measures verified outcomes against the record, never the logs’ self-report.
What is GOVENANT, and who controls it?
An open standard for provable AI agent governance: three laws, a conformance ladder, and acceptance tests any implementation runs against its own record. Free to all under CC BY 4.0, with a citable companion paper (DOI 10.5281/zenodo.21440225), conformance self-assessed and published openly — no certifying authority.

More agent classes

All agent types · All integrations · Run the audit