Forty-eight endpoint types. One policy language.

Connect the subscriptions you already pay for. Ai-Limi normalizes usage, maps rate-limit headers, and lets you move traffic without rewriting the app.

Frontier chat

Highest-quality text, tool calling, and structured output. Workspace and team spend map onto the same cap.

Coverage

Balanced text

Mid-tier chat with prompt cache and batch. Upstream keys stay vaulted; apps receive virtual keys.

Coverage

Long context

Large windows and document ingest. Context cache counts against the same monthly budget.

Coverage

Fast lane

Overflow for latency-sensitive tools. Automatic fallback when a frontier lane returns 429.

Coverage

Open weights

Self-hosted and hosted open models. A cheap default for evals, with a hard cap so they cannot eat production.

Coverage

Regional pin

Keep inference in a declared region. Treat reserved capacity as a prepaid cap, not a mystery line item.

Coverage

Embeddings

Vector writes and retrieval counted in the same ledger as chat, so search cannot hide inside another invoice.

Coverage

Vision & speech

Image, audio, and multimodal calls in a separate pool so experiments cannot starve production chat.

Coverage

Batch & evals

Offline jobs and night runs. Cheap by default, isolated by design, killed at the cap.

Coverage

Need a lane we do not list?

Any chat-completions compatible endpoint connects in minutes. Custom adapters on Scale.

Request a lane