POST /api/zeq/agent/page-chat

LLM page-chat bound to a state machine. Replies stream back with the kernel's actual computed values (operators, R(t), phase, proof_digest) and write a ZeqProof envelope to your audit entangled state. This is what the Pulse calls in slug-bound mode.

Auth

Authorization: Bearer zeq_ak_… OR cookie session OR (for visitor- on-hosted-page) CORS-open with the entangled state's Site SDK key attached server-side. The Pulse on /s/<slug>/ and on third-party embeds works without explicit auth — the framework attaches the key at proxy time.

Cost

25 ZEQ per turn atomic. 80% burns, 20% to foundation. If the slug-bound chain's wallet is short, returns HTTP 402 INSUFFICIENT_BALANCE and falls back to a self-heal message in the Pulse (no LLM call upstream).

Request

curl -X POST https://YOUR-FRAMEWORK/api/zeq/agent/page-chat \
  -H "Authorization: Bearer zeq_ak_..." \
  -H "Content-Type: application/json" \
  -d '{
    "slug": "zeq07090490306",
    "agent_id": null,
    "messages": [
      { "role": "user", "content": "What's the precision of operator KO42?" }
    ],
    "credential_id": null
  }'

Body

Field	Type	Required	Notes
slug	string	yes	Chain to bind the chat to.
agent_id	uuid	no	Bind to a specific spawned agent (uses its ZID + memory). Null → chain-default thread.
messages	OpenAI-shape array	yes	Last 20 are kept; older history is summarised server-side.
credential_id	uuid	no	BYOK credential to use; omit → uses chain's default LLM (Fireworks free or owner's BYOK).
tools_enabled	bool	no	Default true. When false, the LLM cannot call framework tools (compute, contracts, transparency).

Response · 200 OK (non-streaming)

{
  "ok": true,
  "agent_id":  "5b7a9c12-...",
  "reply":     "KO42 enforces ≤0.1% precision against the Zeq metric tensor...",
  "operators_invoked": ["KO42"],
  "compute_results":   [ { "result": {...}, "proof_digest": "abc123..." } ],
  "tally_charge": {
    "charged":         25,
    "burned":          20,
    "toFoundation":    5,
    "remainingAfter":  118
  },
  "x_zeq_compliance_envelope": "env_2026-05-10T13:08:32Z_..."
}

The operators_invoked and compute_results arrays surface what the LLM actually called via tools — the Pulse renders these inline as proof cards under each reply.

Streaming response

Set Accept: text/event-stream to receive SSE chunks (data: {delta: "..."} lines). Useful for long replies; the Pulse streams by default.

Errors

Status	Code	Cause
400	slug_required	Missing slug.
400	messages_required	Empty messages array.
402	INSUFFICIENT_BALANCE	Wallet < 25 ZEQ. Falls back to a self-heal message rendered by the Pulse.
404	slug_not_found	Slug doesn't resolve.
429	RATE_LIMITED	Per-slug quota.

CORS

This endpoint allows wildcard CORS (Access-Control-Allow-Origin: *) because it's the surface the Pulse uses cross-domain. Auth happens via the entangled state's site SDK key, attached server-side at proxy time, so the public origin never sees the key.

Related

• Pulse — what calls this
• Embed the Pulse — how third-party sites get an Pulse
• BYOK — pass credential_id to use your own model
• Agent runtime — sibling endpoints