LiyaEngine Docs
Platform

Intent API

Structured AI actions via a single REST endpoint — no prompt engineering, no model wrangling.

Dashboard

The Intent API is how you invoke LiyaEngine. Pass a domain, an intent key, and your input — get back a validated, structured response. No raw prompts, no system-message wrangling, no parsing free-text output.

Call an intent

POST https://api.liyaengine.ai/v1/run
X-API-Key: liya_prod_...
Content-Type: application/json

{
  "domain": "your-domain",
  "intent": "your-custom-intent",
  "input": { ... }
}

An equivalent path-based form is also available: POST /v1/{domain}/{intent} with just { "input": {...} } in the body.

Response:

{
  "success": true,
  "data": {
    "output": { ... },
    "session_id": "ses_abc",
    "message_id": "msg_xyz"
  },
  "metadata": {
    "intent": "your-custom-intent",
    "domain": "your-domain",
    "model_used": "claude-sonnet-5",
    "tokens_used": 1240,
    "cost_usd": 0.00062,
    "latency_ms": 340,
    "cached": false
  }
}

What makes it different

Intent-native

Call an intent by name — the engine knows the tools to invoke and the schema to validate against.

Structured by default

Every response conforms to your intent's output schema. No parsing, no hallucinated field names.

Domain-scoped

Field names, valid values, and guardrail policies are loaded from your domain config automatically.

Observable

Every call returns a trace: intent resolved, tools invoked, guardrails checked, latency per step.

Intent structure

An intent is a named, schema-bound AI action. It has:

  • Input contract — what fields you must and may pass
  • Output contract — what fields you always get back
  • Tool set — functions the engine can invoke (see Agent Orchestration)
  • Guardrail policies — safety and compliance checks applied (see Guardrails)

You define all of this once, when creating a custom domain — the engine enforces it on every call after that.

What's next?

On this page