ARCHAI Governance API

Three endpoints, one governed request path. Every call to /api/govern/request passes through IDENTARCH, INTENTUM, AGENTUM, MNEMARCH, and ACCOUNTUM before an AI provider is ever reached — and every step is logged to an immutable, hash-chained ledger.

Endpoints

GET /api/govern/status

No auth, no body. Returns the operational state of every mechanism.

curl https://archai.systems/api/govern/status

{
  "status": "OPERATIONAL",
  "version": "2.3.0",
  "chain": [
    { "mechanism": "IDENTARCH", "status": "ACTIVE", "version": "1.0.0" },
    { "mechanism": "INTENTUM",  "status": "ACTIVE", "version": "1.0.0" },
    { "mechanism": "AGENTUM",   "status": "ACTIVE", "version": "1.0.0" },
    { "mechanism": "MNEMARCH",  "status": "ACTIVE", "version": "1.0.0" },
    { "mechanism": "ACCOUNTUM", "status": "ACTIVE", "version": "1.0.0" }
  ],
  "ledger": { "entries": 13, "chainHash": "...", "integrity": "CHAIN_INTACT" },
  "router": "LIVE",
  "storage": "KV_PERSISTENT",
  "ai": "CONNECTED",
  "aiModel": "@cf/meta/llama-3.3-70b-instruct"
}

POST /api/govern/request

The core governance call. Requires a node identity, a justified intent, and the request payload itself.

curl -X POST https://archai.systems/api/govern/request \
  -H "Content-Type: application/json" \
  -d '{
    "node": {
      "id": "agent-01",
      "role": "customer-support-agent",
      "authority": "Acme Corp"
    },
    "intent": {
      "action": "answer-support-query",
      "justification": "Responding to an inbound customer support ticket."
    },
    "request": {
      "prompt": "Summarize this customer complaint."
    }
  }'

Fields

Field	Required	Notes
node.id / role / authority	Yes	Identity anchor for this request — who is asking, in what capacity, under whose authority.
intent.action	Yes	Short label for what's being attempted.
intent.justification	Yes	Minimum 20 characters. Rejected below that.
request	Yes	String or object. If an object, prompt or message is used as the model input.
provider	No	Omit to route to Workers AI. Include to govern your own AI stack instead — see below.

Bringing your own AI provider

By default, governed requests are forwarded to Cloudflare Workers AI. To govern your own OpenAI-compatible endpoint instead, add a provider object:

"provider": {
  "baseUrl": "https://api.openai.com/v1",
  "model": "gpt-4o",
  "apiKeyEnvVar": "OPENAI_API_KEY"
}

apiKeyEnvVar names an environment variable bound in your Pages project settings — the key itself is never sent in the request body.

Response

{
  "success": true,
  "governanceId": "...",
  "version": "2.3.0",
  "governed": true,
  "chain": {
    "IDENTARCH": { "status": "PASSED", "nodeId": "agent-01", "claimId": "..." },
    "INTENTUM":  { "status": "PASSED", "action": "answer-support-query" },
    "AGENTUM":   { "status": "PASSED", "stepsCompleted": 4 },
    "MNEMARCH":  { "status": "PASSED", "memoryId": "..." },
    "ACCOUNTUM": { "status": "PASSED", "sequence": 14, "totalEntries": 14 }
  },
  "ai": { "model": "gpt-4o", "response": "...", "governed": true },
  "message": "Request fully governed and AI response returned."
}

If any mechanism in the chain fails, the request halts there — no downstream mechanism runs, and the AI provider is never called. The response includes which mechanism failed and why.

GET /api/govern/audit

No auth, no body. Returns the most recent ledger entries and chain-integrity status.

curl https://archai.systems/api/govern/audit

{
  "entries": 14,
  "total": 14,
  "chainHash": "...",
  "integrity": "CHAIN_INTACT",
  "ledger": [
    { "sequence": 14, "nodeId": "agent-01", "action": "answer-support-query", "outcome": "SUCCESS", "recordedAt": "...", "entryHash": "..." }
  ]
}

Each entry's hash is computed from the previous entry's chain hash, so any tampering with an earlier entry invalidates every entry after it — that's what integrity: "CHAIN_INTACT" is verifying.

The Five Mechanisms