Platform Overview

Call an intent.
We handle the rest.

Liya Engine is an intent-native AI platform. You send a domain intent; the engine orchestrates agents, retrieval, guardrails, and output validation — and returns structured, production-ready output.

No prompt chains. No tool schemas. No agent loops.

Get started freeRead the docs
< 5ms
Guardrail pre-filter latency
99.95%
Platform uptime SLA
12
Intent endpoints in Talent Pack
< 5 min
First API response from signup
The API surface

Three fields.
Production AI.

Every intent call takes three fields: the domain pack, the intent name, and your input. The engine handles orchestration, retrieval, tool calls, guardrails, and output validation automatically.

The response includes the structured output and a complete execution trace — steps taken, tools called, agent mode used, and latency — so you always know exactly what ran.

packWhich domain to use — talent, compliance, healthcare, legal…
intentThe specific operation — analyze-resume, score-candidate, flag-policy…
inputYour data — resume text, document, query, or structured payload
requestbash
POST https://api.liyaengine.com/v1/run
x-api-key: liya_prod_sk_xxxx
Content-Type: application/json

{
  "pack":   "talent",
  "intent": "analyze-resume",
  "input":  { "resume_text": "..." }
}
responsejson
{
  "response": {
    "content":    "Candidate scores 87/100 for Senior Engineer...",
    "intent":     "analyze-resume",
    "confidence": 0.93,
    "metadata":   { "guardrails_passed": true, "sources": 3 }
  },
  "execution": {
    "steps":       3,
    "latency_ms":  390,
    "agent_mode":  "agentic",
    "tool_calls":  ["fetch_profile", "score_criteria", "rank_fit"]
  }
}
Architecture

Intent in. Structured output out.

Your call hits the Domain Pack's intent router, which maps it to the engine. The orchestrator decides execution strategy, invokes tools from the domain registry, runs retrieval, applies guardrails at every step, then validates output before returning.

Every layer is observable. Every layer is independently configurable on Enterprise plans.

Architecture deep-dive in docs ↗
Your Application
POST /v1/run · { pack, intent, input }
REST / SDK
Domain Pack
Intent routing · Tool registry · Output schema
✓
Agent Orchestrator
Mode selection · Multi-step execution · Error recovery
✓
Retrieval & Memory
RAG pipelines · Session memory · Episodic store
✓
Guardrails + Validation
Policy checks · PII redaction · Output quality pipeline
✓
Foundation Models
Claude · GPT-4o · Gemini · Custom endpoints
↓ execution flows top to bottom · trace returned in response
Platform layers

Everything runs inside the engine — nothing to configure

Each layer is a capability the engine manages on your behalf. On Enterprise, every layer is independently configurable, replaceable, or self-hostable.

◈

Domain Packs

Vertical intelligence, pre-assembled

Each pack bundles the intent schema, tool registry, retrieval config, output schemas, and guardrail policies for a specific industry — versioned and auditable out of the box.

  • ✓Talent, compliance, healthcare, legal, finance
  • ✓8–20 production-ready intents per pack
  • ✓Versioned configs — no silent behaviour changes
  • ✓Custom pack co-engineering for Enterprise
Explore Domain Packs
⬡

Agent Orchestration

The engine behind every intent call

The orchestrator runs inside the engine layer — invisible to callers. It decides whether to use single-shot or multi-step agentic execution, which tools to invoke, and how many reasoning steps to take.

  • ✓Auto mode selection — agentic or single-shot
  • ✓Managed tool registry per domain pack
  • ✓Built-in error recovery and retry logic
  • ✓Full execution trace in every API response
Explore Agent Orchestration
◎

Retrieval & Memory

Context that knows what matters

Hybrid RAG pipelines, semantic chunking, and multi-tier memory inject the right context into each agent step — automatically, without you managing vector stores or prompt windows.

  • ✓Semantic + keyword hybrid search
  • ✓Session and episodic memory per tenant
  • ✓Re-ranking and metadata filters
  • ✓Bring-your-own vector store (Enterprise)
Explore Retrieval & Memory
◬

Guardrails

Policy enforcement at every step

Guardrail checks run before and after each tool call — not just on final output. Domain-specific validators, PII redaction, and output quality pipelines ship with every pack.

  • ✓Pre- and post-tool-call policy checks
  • ✓PII redaction and content filters
  • ✓Output validation and quality scoring
  • ✓Full audit log per intent execution
Explore Guardrails