One gateway for catalog models, every team, every rupee.
unoblox is a full inference control surface — the OpenAI-compatible API your code already speaks, plus routing, guardrails, cost controls and enterprise identity, over a marketplace of providers.
Three compatible surfaces, one gateway
Standardise on whichever request shape your stack already uses. Every surface shares the same keys, routing, guardrails and billing.
Chat Completions
The OpenAI Chat Completions API, unchanged. Point any OpenAI SDK at the gateway and your existing code just works.
Responses
The Responses API surface for apps built on the newer OpenAI response shape — same keys, same routing, same billing.
Messages
An Anthropic-style Messages skin, so teams standardised on that shape can migrate without rewriting their client.
Route by intent, not by endpoint
When several providers serve the same model, unoblox load-balances across their healthy endpoints by price, latency and availability — and fails over automatically. Or hand routing to a virtual model and let the gateway decide.
- unoblox/auto — Balances price, latency and availability to pick the best healthy endpoint for each request.
- unoblox/pareto — Optimises the price/performance frontier — the strongest model that stays inside your cost envelope.
- unoblox/free — Routes to free-tier models where providers offer them — ideal for prototyping and evals.
// Let unoblox choose the best healthy endpoint,
// or name a model directly — same API either way.
const res = await client.chat.completions.create({
model: "unoblox/auto", // or "pareto", "free", or "author/model"
messages: [{ role: "user", content: "Summarise this contract." }],
});Stacked failover means a single provider outage degrades latency, not availability — the request retries the next healthy endpoint.
A catalog you can actually shop
Browse models and providers currently published in the catalog, with the metadata presently available for each row.
Safety you configure, not hope for
- Allow / block lists — Constrain which models and providers a workspace or key can reach.
- Moderation — Screen requests against moderation policy before they leave the gateway.
- Zero-data-retention routing — Route sensitive traffic to endpoints that retain nothing.
- Per-provider data policies — Honor each provider’s data terms, surfaced per model.
Predictable spend, in rupees
Prepaid INR credits
One rupee balance for your whole org — top up with Razorpay or set auto-top-up.
Budgets
Cap spend per workspace and per period; get warned before you hit the ceiling.
Per-key spend limits
Scope every API key to a hard rupee limit so a leaked key can't drain the balance.
Shared credit pool
Teams draw from one org balance instead of juggling separate prepayments.
From your first key to your whole org
Bring your own provider keys, put identity and isolation around your teams, and keep an immutable record of every change.
BYOK — bring your own keys
Already have provider contracts? Route through unoblox for governance, observability and one console while paying providers directly. Keys are encrypted at rest.
Observability
Usage by key, model and workspace; budgets and alerts; an immutable audit trail and maker-checker on sensitive platform changes.
Metadata-only by default
Usage rollups record token counts and cost — never prompt or completion content. Opt into strict ZDR for sensitive workloads.
SAML SSO
Single sign-on with your identity provider.
SCIM provisioning
Automated user and group lifecycle.
Workspaces
Isolate teams, keys, budgets and policy.
Private AI
Dedicated GPUs run for your organisation. Talk to the team.
See the platform for yourself
Review the live catalog and account controls. SSO and BYOK depend on configuration; Private AI: dedicated GPUs run for your organisation. Talk to the team.