Docs and evaluator portal

Inspect the control plane before you trust the claim.

GRIFF AI is the chain-of-custody layer for agentic work. Start here to run the demo, verify proof, import the public schema, read the standards map, install PromptGate, and review the current security posture.

Quick paths

What a serious evaluator should open first.

Request controlled packet

Evaluator checklist

Public demo acceptance criteria.

A demo is not procurement evidence unless the buyer can inspect the contract, see the policy decision, verify the chain, and understand what was blocked. These are the fields every public proof surface should expose or link to.

Task contract

Objective, target resources, allowed actions, forbidden actions, autonomy mode, and risk tier.

Policy decision

Allow, deny, request changes, approval requirement, and rationale.

Approval state

Named human approval for privileged or sensitive work.

Execution evidence

Runner name, environment, artifact, secrets touched, destructive operations, and result.

Hash-chain proof

Previous hash, event hash, chain head, tenant id, and verification result.

Export path

Buyer-safe certificate, audit pack, OpenAPI schema, and controlled packet request path.

Platform tracks

How the public site maps to product work.

MASTER ATC

The flagship custody system: agent identity, tasks, tools, policies, approvals, replay, and export.

PromptGate

Public preflight wedge: prompt audit, score gate, repeat/stop controls, and novice LLM install.

CostGuard

Planned token-spend and model-routing lane for provider costs, cache hits, anomalies, and owner alerts.

Recall and Memory

Protected recall service with provenance, supersession, policy gates, and public OpenAPI mirror.

Evidence Lake

R2-backed artifacts with D1 metadata, signatures, chain heads, and exportable procurement proof.

Developer Portal

This route becomes the stable public map for SDKs, OpenAPI, webhooks, examples, and proofs.