Build AI you can trust into anything.

Authentication: Bring Your Own Key

ASURIQ uses BYOK (Bring Your Own Key) authentication. You provide your own LLM API key from any supported provider. ASURIQ never stores your keys server-side. They're used for the duration of the request only.

Quick start

curl -X POST https://consensus-production-6eeb.up.railway.app/api/v1/check \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Your AI response text here",
    "query": "The original question",
    "tier": "protocol"
  }'

Try it live

Pre-loaded with a classic AI hallucination. Hit send to see what ASURIQ catches.

Verify API

Database-backed AI answer verification. Checks claims against 21 real-world databases. Not AI checking AI. No authentication required.

Cognitive Stack API

11-system cognitive architecture exposed as API endpoints. Multi-perspective deliberation, adversarial reasoning, epistemic calibration, reflection, and self-improvement. All endpoints are BYOK.

Code Quality API

Structural code review, test health assessment, mutation testing, codebase analysis, UX validation, architecture review, and design system audits. All BYOK.

Routing API

Intelligent model selection across providers. Classifies query complexity, selects the cheapest adequate model, proxies the call, and tracks savings. Single-provider and multi-provider panel consensus mode.

Memory API

Per-user vector memory with hybrid recall (vector + keyword + quality-weighted RRF merge). Includes entity graph, automatic NIGHTSHIFT maintenance, and 10K observation cap per user. Requires Memory add-on subscription.

Hands API

The action layer: HTTP proxy, web extraction, scheduling, email, file operations, shell execution, and semantic search. ASURIQ gives judgment, Hands gives action.

WorkForce API

Multi-model deliberation engine. Submit a decision, get a bid with estimated credit cost, approve it, then execute. Convergence-driven with parallel tier execution, cross-contextualization, and deadlock detection.

Quality gates

Some endpoints require a minimum model tier. If your model is below the threshold, the endpoint returns a structured quality_gate response instead of degraded output. It tells you exactly what tier you need and why.

{
  "result": null,
  "quality_gate": {
    "triggered": true,
    "tool": "cognitive/challenge",
    "minimumTier": "balanced",
    "detectedTier": "fast",
    "detectedModel": "Haiku",
    "explanation": "challenge requires a balanced-tier model or higher. Adversarial reasoning requires nuance.",
    "whatYouWouldGet": "Steel-manned counterarguments with evidence chains.",
    "upgradeOptions": [
      { "action": "Switch to a balanced model", "how": "Change to Sonnet, GPT-4o, or similar.", "cost": "Free" },
      { "action": "Use ASURIQ Routing", "how": "Subscribe to Routing BYOK and let ASURIQ select the right model.", "cost": "$13/mo" }
    ]
  }
}

Credit billing

Credit-consuming endpoints (Verify detail, Routing managed, Hands hosted, WorkForce) use atomic deduction with automatic refund on failure. Buy credits in packs. Larger packs cost less per credit.

{
  "error": "Insufficient credits: 2 available, 10 required",
  "remaining": 2,
  "required": 10
}

API pricing

Subscriptions for zero-marginal-cost products (IP delivery, local execution). Credits for compute-backed actions (server execution, database lookups, LLM inference).