White paper · Charter · October 2026
Your Knowledge. Your Rules. Proof.
How Charter gives AI agents your company's approved knowledge and rules, and keeps the record of what governed their work.
Inside this paper
- The problemEach agent gets the rules separately, and the copies drift.
- How Charter worksEquip. Manage. Prove., in four steps on each piece of work.
- The architectureCanon, SIGN™, Trellis and MCP, each with one job.
- What Charter claimsAnd what it does not.
The rule gap, measured on ourselves. A self-audit of about 30 of our repositories found 412 business rules written into application code. Checked against Canon, 31% were already fully governed, 24% were partly covered, and 45% existed only in code (internal).
The problem: each agent gets the rules separately
Models know the world. They do not know your company. Most companies fill that gap one agent at a time: the rules go into a prompt, a project, a retrieval index or a block of code, and the next team does it again. Each copy drifts from the others and from the policy it came from, and no one can say later which version an agent used.
Gartner predicts that by 2028 the average global Fortune 500 enterprise will have over 150,000 agents in use, up from fewer than 15 in 2025. A rule that lives in each agent has to be maintained in each agent.
How Charter works: Equip. Manage. Prove.
Charter is the enterprise authority layer for AI. It moves the company's knowledge and rules out of the prompt and into a governed layer. The agent draws on that layer while it works, and the layer keeps the record.
Current company knowledge
Approved business rules, standards, policies, definitions, procedures and expert context.
How the agent operates
Purpose, identity, access, actions, approvals, guardrails, escalation and operating status.
The evidence
The source, rule, version, decision, approval and action that governed the work.
On each piece of work Charter determines what applies, makes the governance decision, governs the action through the governed path, and creates the record of what governed it.
Why the layers you already own are not enough
Retrieval, identity, observability and agent frameworks each answer a different question. None of them establishes, on the company's behalf, which approved rule applies to this piece of work.
| Layer | What it does | What Charter adds |
|---|---|---|
| RAG and enterprise search | Finds relevant documents and passages | The authoritative source, current version, applicable rule and decision evidence |
| IAM | Authenticates identity and grants technical access | Business authority, conditions, approvals and governing knowledge |
| Observability | Records prompts, calls, costs and failures | The authoritative resolution behind the trace |
| Agent frameworks and orchestrators | Plan and execute agent workflows | Enterprise-controlled knowledge, independent of the framework |
The architecture
Charter is the system. Four components sit under it, each with one job.
Canon
The governed source for enterprise knowledge and business rules.
SIGN™
An open notation agents can read. Open, Apache 2.0.
Trellis
The control plane. Its WARRANT capability produces the replayable receipt.
MCP
The open protocol approved agents use to connect.
LLMs are probabilistic. The system around them does not have to be. Charter makes the inputs to the work governed and reproducible. It does not make the model's output deterministic, and it does not declare legal or regulatory compliance on a customer's behalf. The full paper sets out every claim boundary, the worked example, the evidence from running our own company on Charter, and where to start.
Read the full paper
The complete paper: the rule gap, the four steps, a worked rule change, what existing layers do and what Charter adds, the architecture, the claim boundaries, the evidence, and six questions for a first use case.
Related
How Charter works
Why agents need a governed source, what Charter adds to retrieval, IAM and observability, and how Canon, SIGN™, Trellis and MCP fit together.