AveryRulebook

The compliance platform

From policy and regulation to a gate every agent must pass.

Rulebook is a compiler, registry and compliance decision runtime. It gives every agent one place to check what is allowed, prohibited, required or uncertain before it answers or acts.

One compliance lifecycle

Authoritative requirements in. Policy-governed agent actions out.

Every rule remains linked to source evidence, reviewer confirmation and a published version. Each governed decision creates an auditable trace that remains reproducible even as source material changes.

DECISION / RB-204817 MS
Agent asksCan this customer record be shared with the service partner?
SourceData handling policy 7.4
ContextEU customer · Support purpose
DecisionALLOW WITH OBLIGATIONS
Governed answer

Yes. Share only through the approved subprocesser after direct identifiers are redacted. Retain evidence of the transfer for 24 months.

Decision trace4 checks
  1. 01
    Purpose limitation satisfiedPolicy 7.4.1 · support delivery is an approved purpose
    PASS
  2. 02
    Regional control appliedGDPR Art. 5(1)(b) · EU customer context detected
    APPLIED
  3. 03
    Partner status verifiedVendor control VR-12 · subprocessor is on the approved list
    PASS
  4. 04
    Disclosure minimizedData handling policy 7.4.3 · identifiers must be redacted
    REQUIRED
01

Use the approved subprocessor

02

Redact direct identifiers

03

Retain transfer evidence for 24 months

Policy snapshotrulebook/data-handling@7.4.32026-08-02
Signed receiptsha256: 7e42a1...91bcVERIFIED

The compliance control loop

Compile. Reason. Enforce. Prove.

Rulebook turns policy and regulatory material into confirmed rules, applies the lowest sufficient reasoning tier, gates the answer or action, and produces audit evidence in the same flow.

Rulebook compliance loopPolicy and regulatory rules become an action-time gate
Published version active
01

Compile

Turn sources into proposals

  • Policies and procedures
  • Contracts and spreadsheets
  • Repositories and standards
Typed rules with source evidence
02

Reason

Climb only when needed

  • T0 deterministic
  • T1 bounded judgment
  • T2 replayable plan
  • T3 labelled explanation
A declared answer tier
03

Enforce

Decide before the action

  • Allow or deny
  • Require approval
  • Return obligations
  • Transform the request
A binding compliance decision
04

Prove

Carry the evidence forward

  • Signed decision receipt
  • Append-only access ledger
  • Independent offline verification
A portable proof record
Human confirms Nothing self-promotesAnswer declares Tier and reproducibilityEvery call Writes a ledger entry

How it works

From source material to a published compliance gate.

01

Connect the requirements

Bring policies, regulations, contracts, spreadsheets and code that define the enterprise's compliance obligations.

02

Compile proposals

Rulebook extracts explicit obligations, conditions, citations, precedence and unresolved conflicts.

03

Confirm and test

Authorized policy, legal and compliance experts inspect proposed rules, run cases and publish an immutable version.

04

Check every decision

Agents call REST, MCP, CLI or SDK interfaces with the context of the proposed answer or action.

05

Gate the action

The gateway enforces allow, deny, obligations or human review before protected tools execute.

06

Verify the audit trace

Every governed outcome can include a signed record tied to inputs, evidence and the published rules that controlled it.

The answer ladder

Use the least intelligence needed.

The caller sets the ceiling. Rulebook climbs only when the tier below cannot answer, never to produce a more fluent response.

T0 · Decided

Symbolic resolution

No model call. The same published version, facts and time produce a bit-reproducible decision path.

Bit-reproducible
T1 · Decided with judgment

Bounded oracles

Approved scalar judgments can narrow or evaluate a condition. Every judgment is recorded verbatim.

Reproducible with recorded judgments
T2 · Reasoned

Validated plan

A model proposes a closed, typed plan. Rulebook validates and executes it without model, network, filesystem or clock access.

Replayable from the stored plan
T3 · Explained

Labelled narrative

An entailment-checked explanation is composed over the T0 to T2 result. It is labelled and never changes the outcome.

Narrative is not reproducible

Human review is an outcome, not a reasoning tier. Missing facts, ambiguity, exceptions or configured policy can require the named authority before an action proceeds.

Compliance control properties

Designed for enforceable, auditable enterprise policy.

  • Unsupported conclusions are not representable as published facts
  • Compiled policy and regulatory proposals require human confirmation
  • Published compliance-rule versions are immutable
  • Model use is visible in the decision record
  • Signed receipts can be verified independently
  • Policy interfaces remain model and agent independent

Fits the existing estate

One compliance decision API across every agent stack.

Use REST, MCP, CLI and language SDKs to place Rulebook at response, tool, approval and workflow boundaries.

Explore integrations

Executive briefing

Put a compliance gate before every consequential AI action.

Bring one consequential agent workflow. We will map the governing policies and regulations, runtime gate, human authority and auditable evidence path with your team.

Request a briefing