Resolve

The resolve endpoint maps natural language queries to matching agents.

Resolve Intent

POST /resolve

{
  "query": "book a flight to Tokyo",
  "topK": 5,
  "protocol": "rest"
}

Field	Type	Required	Description
query	string	Yes	Natural language intent
topK	number	No	Max results (default: all)
protocol	string	No	Filter by protocol

Response:

{
  "status": "ok",
  "results": [
    {
      "passport": {
        "id": "agt_flight_master",
        "displayName": "FlightMaster Pro",
        ...
      },
      "score": 87.5
    }
  ],
  "intent": {
    "intent": "travel_booking",
    "confidence": 0.92,
    "capabilities": ["search_flights", "book_flight"]
  }
}

Parse Intent (Preview)

Parse intent without resolving to agents:

POST /intent/preview

{
  "query": "scan for security vulnerabilities"
}

Response:

{
  "ok": true,
  "intent": "vulnerability_scan",
  "capabilities": ["scan_network", "scan_code"]
}

Scoring

Agents are ranked by composite score:

1. Capability match (0-40) — Keyword overlap with capabilities
2. Domain match (0-20) — Exact or inferred domain match
3. Description relevance (0-15) — Fuzzy match on description
4. Metadata bonus (0-5) — Higher version agents score better