Neural Network Navigator
Integrate AI without rebuilding every provider connection
One authenticated API in front of the aggregated catalog. Route by objective, keep one usage ledger, and add models without shipping a new integration.
Interface
Route by objective
Ask for an outcome instead of naming a model. The router applies your workspace policy, permitted providers and region rules, then returns the model it used together with the exact usage record.
POST /v1/route
Authorization: Bearer $DINNNER_API_KEY
Idempotency-Key: 4f0c1e8a-...
{
"objective": "quality",
"modality": "text",
"region": "eu",
"input": "Summarise this contract clause."
}
200 OK
{
"id": "req_01J...",
"model": "frontier-general",
"provider": "provider-a",
"usage": { "input_tokens": 1180, "output_tokens": 240, "cost_usd": 0.0083 }
}Documentation scope
What is published for integration teams
| Area | Published detail |
|---|---|
| Authentication | Project-scoped API keys with role separation and rotation. |
| Supported resources | Chat, completions, images, video, speech, embeddings, workflows. |
| Asynchronous lifecycle | Job submit, poll, cancel and terminal states for media work. |
| Idempotency | Client-supplied idempotency keys on every mutating request. |
| Rate limits | Per key and per resource limits with explicit headers. |
| Webhooks | Signed deliveries, replay protection and documented retry schedule. |
| Usage reconciliation | Per-request usage records that match your invoice lines. |
| Errors and refunds | Documented error classes and credit handling for failed work. |
| Data handling | Declared processing roles, regions and retention behaviour. |
| Changelog | Versioned catalog and behaviour changes with dates. |
Request API access
Write to us with your company and website, the use case, the countries your users are in, expected volume and the capabilities you need. We confirm fit, permitted regions and commercial terms before issuing keys.