Control AI traffic in production
One integration for your model providers, with policies, budgets, fallback and a trace of every request.
- Task mix
- extraction 0.7 · summarisation 0.3
- Preset
- Balanced
- Router
- Artemis
| Candidate | Expected score | Result |
|---|---|---|
| Model A | 0.93 | selected |
| Model B | 0.91 | eligible |
| Model C | 0.84 | eligible |
| Model D | — | excluded: region policy |
What you control
- Provider access
- One endpoint. Your provider keys or ours.
- Reliability
- A provider fails, the next best one takes over.
- Spending
- Monthly budgets with hard limits per project.
- Governance
- Allow-lists, region rules and PII redaction.
Your app
One request
model: "auto"
NuRoute Gateway
Step 1
Policy and budget checks
Allowed models, regions, PII redaction and spend limits are applied first.
Step 2 · Artemis by NuRoute
Artemis selects the model
It profiles the task and picks the model with the best expected quality for the cost.
Step 3
Provider and fallback
The request goes to an allowed provider. If it fails, NuRoute tries the next best.
Request trace. Every decision, with the candidates considered, is logged and can be replayed.
Providers
Your keys or NuRoute's
Commercial, European and self-hosted endpoints.
When something goes wrong
- A provider fails
- Next best model and provider, up to 3 tries
- Policy excludes every provider
- Refuse, or relax the rules. Your setting
- Caps rule out every model
- Refuse, or a simpler fallback decides
- Every provider is rate-limited
- Returns a 429
Where Artemis fits
With model: "auto", Artemis is the model-selection step, after your policies. Name a model instead and everything else here still applies.
Not yet
- Streams are not retried across providers
- Budget alerts are not sent yet
- The audit trail is partial: decisions are logged and replayable
Need it inside your own infrastructure?
See Enterprise