Browse all documentation
Agent platform · 5 of 9

Intent routing

Rank eligible agents with a versioned, inspectable routing receipt.

betarouting receipt v1 · server 3.0Verified 2026-07-29Owner: ENS.Tools Agent Routing
Open this feature

The intent router maps an outcome to eligible ENS agents without a hidden model score. Capability matching and hard constraints run first. The response binds the normalized request and candidate evidence into a versioned receipt that can be stored for audit and semantic-drift detection.

Natural-language intent matching against advertised context and capabilities.
Alias-aware, deterministic capability matching required by default.
Protocol, minimum-health, freshness, verified-only, and minimum-score hard filters.
Capability 45 + health 20 + verification 15 + recency 10 + latency 10.
Request hash, evidence digest, stable route ID, matched concepts, and every score component.
Explicit exclusion reasons and counts; constraints are never silently relaxed.
No model-generated relevance score.
Production contract

Before you run this workflow

Networks

  • No transaction network; routes use indexed Ethereum ENS evidence

Prerequisites

  • A concrete intent
  • Indexed candidates with sufficiently fresh evidence

Wallet permissions

  • None

Expected result

  • Versioned request hash, route IDs, evidence digest, score components, and exclusion reasons

Errors and recovery

  • Inspect exclusions before changing constraints
  • Refresh candidate health
  • Alert on fixed-fixture semantic drift
ENS.Tools agent intent router ranking four ENS agents with a visible score breakdown
The router shows matched terms and every scoring component instead of a hidden model ranking.
Workflow

How to use it

  1. 1

    Describe an outcome

    Use a concrete task such as “resolve an ENS identity and inspect its portfolio.”

  2. 2

    Apply constraints

    Choose a required protocol, minimum health, maximum evidence age, verification requirement, and minimum score before ranking.

  3. 3

    Inspect the receipt

    Persist routingVersion, requestHash, routeId, evidenceDigest, matched concepts, exclusions, and every score component. A high score is relative to bounded candidates, not a universal endorsement.

  4. 4

    Detect semantic drift

    For fixed fixtures, alert if routes, exclusions, or scores change unexpectedly even when the API still returns HTTP 200.

  5. 5

    Re-check before execution

    Open the agent profile and refresh endpoint health before privileged or expensive work.

  6. 6

    Use the machine contract

    Read /.well-known/agent-skills/directory-routing.json and call POST /api/agents/route or route_agent_intent with requireCapabilityMatch=true.