Agent governance control plane
Prove what governed
this decision on the 14th.
Your agents run where they live. RuleKeeper holds the documents that govern them and the record of what applied.
The panel is the shape of a run record. The document names and versions are tenant one's governance set.
Governing documents in force at this run
- system-prompt1.0.0
- task-instructions1.0.0
- behavioral-constraints1.0.0
- tool-use-policy1.0.0
- escalation-paths1.0.0
The question
One of these questions has tools. The other has a deadline.
- Asked in engineering
- Is this prompt any good? Trace the run, score the output, compare two versions. Real work, well served by developer tooling, and entirely a development question.
- Asked in an examination
- What governed this decision on the 14th? Who wrote that rule, who reviewed it, what proved it was safe before it went live, and where is the version that was in force that day? That is a compliance question, and nobody serves it.
RuleKeeper is built for the second question, about a date that has already passed, answered under someone else's deadline.
Built for regulated operations deploying agents that take actions rather than make suggestions.
- Mortgage servicing
- Insurance claims
- Healthcare revenue cycle
- Banking operations
- Title and escrow
What it holds
Four records, and every one of them is dated.
Agents keep running where they already are. What arrives here is the record of what they were told and what they did about it.
Agent registry
Every deployed agent with its owner, its stated purpose, what it costs to run, and the escalation rate it committed to when it was registered. An agent whose real rate drifts from its declared rate is a change to the business, and it shows up as one rather than being discovered later.
Governing documents
Five documents per agent, each versioned, each carrying an author, a reviewer and the promotion event that put it in force. Every prior version is kept, because the question in March is what the document said in January.
- system-prompt
- task-instructions
- behavioral-constraints
- tool-use-policy
- escalation-paths
Evaluation harness
Fixtures with expected outcomes, scored against one specific draft. A run that did not finish is not a result and cannot stand in for one, which is the difference between a gate and a formality.
Runs and steps
Every run stamps the exact version of every governing document in force when it happened, with the tool calls, the escalations and the refusals underneath it. Not the prompt as it reads today. The one that was actually applied.
The gate
A change to a rule is a draft until the evidence clears it.
Editing a prompt changes behaviour everywhere at once and there is no compiler to catch it. So the edit is not the change. The promotion is the change, and it has to be earned.
- 01
Draft
The edit is written as a draft. It governs nothing: no run reads it, no agent behaves differently, and it can sit there for a week while the reviewer is on leave.
- 02
Score
The eval suite runs against that exact draft. Not against the current version, not against a copy that has since moved. Only a finished run counts as a score.
- 03
Promote
Promotion is refused unless a finished run scored that draft at or above the standing baseline. Refusal is the default, and it names the fixtures that broke rather than reporting a number.
- Agent
- allowance-classifier
- Document
- behavioral-constraints
- Draft
- 1.1.0
- In force
- 1.0.0
Scored on the draft
9 / 11
standing baseline 11 / 11
Below the baseline, so the draft stays a draft. Nothing in production changed.
Fixtures the draft broke
escalate-out-of-season-winterize-tx
Winterization ordered in Texas outside the state season. The agent is required to refuse the classification and escalate. Under the draft it classified confidently.
hawaii-winterization-not-required
Winterization ordered in a state that has no winterization season at all. Same requirement, same failure.
Tenant one relaxed the behavioral constraints on one agent deliberately, to check the gate would fire. It did. The draft is still on record as withdrawn, which is a different thing from superseded, and the difference is visible in the document history.
Conformance
The failures that do not announce themselves.
None of these produce a bad output, an error or an alert anywhere else. They are structural, they are true for months, and they are only visible from a registry that holds every agent at once.
- grant.irreversible.no_gateAn irreversible tool grant with no escalation gateA tool that files, submits, pays or deletes, held by an agent that can reach it without a human in the path. The grant and its side effect class are both in the registry, so this is a query, not a review.critical
- agent.no_eval_baselineAn agent with no eval baselineIn production with nothing to regress against. Every future change to its governing documents is unmeasurable, and the first sign of a problem is the incident.warning
- agent.no_declared_rateAn agent that declares no expected escalation rateWithout a declared number there is no such thing as drift. An agent escalating on two runs in a thousand is either working correctly or was never wired to a gate, and nobody can tell which.warning
Proof
One tenant. Saying so is stronger than implying more.
4
Governed agents
Each with a declared escalation rate
20
Governing documents
Five per agent, every version kept
46
Eval fixtures
All currently passing
2
Fixtures caught the degradation
Promotion refused, draft never landed
Tenant one is a live mortgage field services system. Its four agents classify preservation allowances against published HUD guidance, assess denied claims, triage contractor field reports, and detect the vacancy exposure windows that turn into disallowed costs later.
There are no other customers yet, no logos to show and no industry benchmark to quote. What there is, is one system where every control described on this page is switched on, and a regression history that includes degradations introduced on purpose, because a gate nobody has watched refuse anything is not a gate.
Open it
Bring the date. The record is already there.
Agents keep running in your own application. What changes is that six months from now, the answer to what governed a decision is a row rather than a reconstruction.