Route each AI request by cost, sensitivity and reliability

Not every call needs an expensive model — and not every call should leave your premium provider. Pick the right mode per workflow. Cheap mode is live today; Balanced and BYOK are rolling out.

Cheap

Available now

Decentralized open-model inference at the lowest cost. For public, synthetic and non-sensitive workloads.

Best for

SEO, content, product descriptions, public docs, summaries, tests, staging.

Best-effort availability. Do not send secrets or regulated data.

Balanced

Rolling out

Cheap first, with automatic fallback to a premium provider on timeout, error or latency threshold.

Best for

Client-facing bots where cost matters but failures hurt UX.

In private rollout for founding agencies.

Premium / BYOK

Rolling out

Your own provider keys for sensitive clients, keeping the Mingles dashboard, limits and analytics.

Best for

Private data, production support, regulated workflows, high-value calls.

In private rollout for founding agencies.

Data-sensitivity guide

A rule of thumb for which mode fits which data. When in doubt, route up.

Data type Recommended mode
Public docs Cheap / Balanced
SEO / content Cheap
Product descriptions Cheap
Synthetic data Cheap
Customer PII Premium / BYOK
Legal contracts Premium / BYOK
Private code Premium / BYOK
Regulated data Do not use cheap mode

Cheap mode runs on a decentralized inference network. Treat it like a public, best-effort tier — never send secrets, customer PII or regulated data through it.

Fallback policy examples

The shape of a routing policy. Balanced and BYOK enforcement are rolling out for founding agencies — we set these up with you today.

if timeout > 8s            → fallback to premium
if error                   → retry once, then fallback
if prompt contains PII     → premium / BYOK
if workflow = batch_content → cheap
if workflow = client_support → balanced or premium

Design a routing policy for your agency

Send us one workflow. We’ll map sensitive vs non-sensitive calls and show what can route cheaper.