Private design-partner pilot

Apply one approved retrieval policy across every crawler and agent.

Check each proposed retrieval against your policy and the evidence your integration supplies. Get proceed, constrain, license, review or block—with reasons and constraints your agent must enforce.

Explore a demo Connect an agent Run a measured pilot

Decision boundary: AgentGuard reports an operational result from your configured policy and the evidence supplied to it. It does not grant a license, determine legal rights, or replace legal review.

Declared request

Agent: research-crawler-prod
Purpose: summarize
Retention: 90 days
Output: attributed

policy_0183 evidence claims

Final decision

Conditional

engine v0.1.0

Preserve source attribution

Delete source content after 30 days

sample robots sample RSL sample customer policy

02Evaluation stages

05Operational outcomes

<100msWarm-path design target

100%Versioned decision target

Target retrieval workflow

One checkpoint. Two honest stages.

The production roadmap separates acquisition from response use because a response can carry narrower signals. Today, the pilot evaluates normalized evidence supplied by the customer; it does not collect the target resource.

Acquisition preflight

Evaluate the proposed resource, declared purpose, caller-supplied normalized evidence and tenant policy.

Pilot output: proceed · deny · review

Quarantined observation

A planned customer-side enforcement point can capture response headers and safe metadata before releasing content downstream.

Roadmap: content remains outside agent context

Response-use decision

The target workflow combines response evidence, caller-attested entitlement records and stricter tenant constraints.

Roadmap: release · constrain · licensing review · block

Operational, not rhetorical

Five results your integration can map to actions.

Every API result includes stable reason codes, policy revision, considered evidence, freshness and the evaluator version.

Explore the decision model

01Proceed

Your configured policy returns no blocking condition on the supplied evidence.

02Conditional

Your integration should apply the returned retention, attribution or output limits.

03License

Supplied policy evidence requires a separate customer licensing review.

04Deny

Your configured policy blocks acquisition or downstream use.

05Review

Ambiguous evidence is held for your designated policy owner.

Built for the retrieval path

Replace scattered exceptions with one policy decision point.

Normalize the evidence

Keep robots rules, usage signals, response headers, RSL terms and customer observations separate—with provenance intact.

Evaluate entitlement records

The pilot treats entitlement data as customer-supplied evidence. Registry-backed verification is planned; an allowlist alone never creates rights.

Validate before enforcement

Start with sanitized replay traffic. Live shadow or enforcement depends on a separately scoped customer integration and is not a public-site feature.

Explain every result

The evaluator returns its version, policy revision, considered evidence IDs and reason codes. The current API accepts no page bodies.

Source-handling roadmap

Target architecture: on demand by default, selectively cached when traffic repeats.

AgentGuard is not building or claiming a continuously current permission database of the internet.

Current pilot

The caller supplies bounded, normalized evidence and entitlement records. The API evaluates them without requesting the resource URL.

On-demand collection · planned

For new or expired resources, an isolated collector would request only relevant published policy signals when a customer asks.

Selective TTL cache · planned

Repeated domain/path signals may be reused only within a source-specific freshness window, with the evidence snapshot tied to each result.

Pre-warmed source packs · optional

Enterprise customers may nominate high-value domains for scheduled refresh. No whole-web crawl or full-page archive is planned.

Designed for production integration

A small call in the path. An explicit result back.

The current authenticated pilot API evaluates normalized evidence supplied within an agreed pilot scope. The public sandbox remains fixture-only, and neither path fetches a target URL.

Strict request and response contract
Direct deterministic evaluator integration
Stable reason codes and fail-closed outcomes
Server-owned evaluation timestamps

example-response.json

{
  "schema": "agentguard.decision.v1",
  "decisionScope": "CUSTOMER_CONFIGURED_POLICY",
  "notLegalAdvice": true,
  "receiptState": "NOT_ISSUED",
  "decision": "PROCEED_WITH_CONSTRAINTS",
  "stage": "RESPONSE_USE",
  "enforcement": "ALLOW",
  "reasonCodes": [
    "USAGE_CONSTRAINED",
    "CUSTOMER_POLICY_CONSTRAINT"
  ],
  "constraints": {
    "maxRetentionSeconds": 2592000,
    "attributionRequired": true
  },
  "policy": { "revisionId": "policy_018" },
  "evaluatorVersion": "agentguard-policy-engine/0.1.0",
  "validUntil": "2026-09-04T13:00:00Z"
}

Now recruiting design partners

Prove it on your traffic before you buy a platform.

Start with a bounded replay pilot using a sanitized retrieval sample, your policy rules and an engineering owner. Live shadowing requires a separately scoped integration.

Discuss a pilot See pilot pricing