// NATURAL LANGUAGE → ORCHESTRATED ACTION

One SDK call.
Infinite agents.

IntentKit translates what users say into what agents do. Pass in natural language, get back orchestrated results. The interface to complete the task appears on its own.

                
                app.js
            
                const response = await fetch('https://intentkit.polsia.app/api/resolve', {

                  method: 'POST',

                  headers: { 'Content-Type': 'application/json' },

                  body: JSON.stringify({ input: "What's the weather in Tokyo?" })

                });

                const result = await response.json();

                // IntentKit parsed the intent, selected the right agent,

                // and returned structured results.

                console.log(result.agent);      // "weather"

                console.log(result.action);     // "weather_lookup"

                console.log(result.data);       // { location: "Tokyo", current: { temperature_f: 72, ... } }

                console.log(result.resolved_in_ms); // 847

// TRY IT LIVE

Playground

Type anything. IntentKit parses your intent and routes it to the right agent in real time.

Weather in SF 15% of 2400 sqrt(144) London forecast Quantum computing Math expression

Response

Results will appear here...

// HOW IT WORKS

From words to results in four steps

Intent Parsing

Natural language input is decomposed into a structured intent schema. Entity extraction, action classification, and context resolution happen in milliseconds.

Agent Selection

The routing engine matches the parsed intent to the best available agent. Model-agnostic: works with any LLM provider, any tool, any API.

Orchestration

The selected agent executes the task. Multi-step workflows are handled automatically. Fallback agents activate if the primary fails.

Structured Response

Results return as structured JSON with the parsed intent, selected agent, execution data, and timing. Ready to render in any UI.

// API REFERENCE

Three endpoints. That's it.

POST /api/resolve Parse intent & route to agent

// Request { "input": "What's the weather in Tokyo?" } // Response { "success": true, "intent": { "action": "weather_lookup", "entities": { "location": "Tokyo" }, "confidence": 0.95 }, "agent": "weather", "data": { "location": "Tokyo", "current": { "temperature_f": 72, "condition": "Partly Cloudy" } }, "resolved_in_ms": 847 }

GET /api/agents List available agents

// Response { "agents": [ { "action": "weather_lookup", "name": "weather", "description": "Looks up weather conditions..." }, { "action": "calculate", "name": "calculator", "description": "Evaluates math expressions..." }, { "action": "web_search", "name": "web_search", "description": "Searches the web..." } ] }

GET /api/stats Usage statistics

// Response { "stats": { "total_resolves": 1247, "avg_resolve_ms": 623, "last_24h": 89, "by_agent": [{ "agent": "weather", "count": 482 }, ...] } }

// DESIGN PRINCIPLES

Built for developers who ship

</>

One Integration

Stop wiring together LangChain, vector stores, agent frameworks, and UI libraries. One API. One call. Done.

↔

Model Agnostic

Routes to the best model for each task. OpenAI, Anthropic, open-source. You define the agents, IntentKit picks the right one.

{ }

Intent Schema

Structured, versionable, composable intent definitions. Your schema becomes smarter over time as IntentKit learns from usage patterns.

■

Always Structured

Every response is typed JSON. No string parsing. No guessing. Pipe it straight into your components, your database, your pipeline.

The last SDK you integrate.

Every app becomes an intent-driven experience. Every user request finds the right agent. Every result gets the right interface. This is intent-based computing.

One SDK call. Infinite agents.