Inference & APIs

Every model. One API.

160+ models from every major provider, behind one OpenAI-compatible API. Point your existing code at our base URL and it works.

Run it direct. Or run it guarded.

Every request goes one of two ways. You choose, per request.

provider prices, nothing added

The same models, at the price the provider charges. We add nothing.
Built for pipelines, batch jobs, agents and internal tools: anywhere you need fast, reliable model access with failover already underneath.

Provider price + platform price for guarding.

The same request, passed through the trust layer before and after the model: safety checks tuned for real human moments, alignment to your values, moderation of what comes back, citations when grounded.
Built for anything your users will see: chat experiences, member care, youth-facing tools, and especially moments of care, crisis, or conviction.

CAPABILITIES

Failover built in
direct providers first, health-checked fallbacks behind them, automatic recovery.
Prompt caching
cache-aware billing with TTL you control.
Grounded completions
answers cited to your sources, structured for your UI.
Web search, cited
current answers with URL citations, server-side.
Multimodal
vision in, image generation out, tools throughout.
Streaming everywhere
SSE with keepalives and truncated-stream detection.

Guarded is the trust layer: three checkpoints on every request, before, at, and after the model. Direct skips all three.

Before the model
every request is screened by Gloo's guardrails, and unsafe requests are refused.
At the model
Gloo's values alignment shapes the response, and tradition settings tailor it to your context.
After the model
the response is moderated as it streams, and an unsafe answer is replaced before your user sees it.

MODEL CATALOG

New models land behind the API you already integrated. No rewrites, no migrations, no keeping up.

Some of the labs behind the catalog.

166 models in the catalog today, every one behind the same API.

FAQ

THE REST OF THE PLATFORM