Providers
Connect the accounts you already pay for.
INFRO runs your requests on your own provider accounts. You keep the contracts, the negotiated rates, the quotas and the invoices — and the gateway adds routing, failover, request traces, cost attribution and policy around them. It takes no part of what you pay a vendor.
INFRO is in private early access. The catalog, the docs, and the API contract are public today, and the gateway is deployed. The console opens with an early-access account, billing opens in stages, and anything beyond v1 is marked Later. Early access details
Your provider accounts.Your routing rules.
Connect the providers your organization uses. Choose models directly or send requests through a route you configure.
Provider connections
Sample dataYour organization's own provider accounts.
Production route
Sample dataThe order a request tries your providers.
- 1. PrimaryOpenAIopenai-prod
- 2. Fallback 1Anthropicanthropic-prod
- 3. Fallback 2Googlegoogle-prod
Requests try OpenAI first. If it fails with an eligible error or a timeout before the response starts, the request moves to Anthropic, then to Google.
Named route
Uses the targets you configure.
Direct model
Uses an eligible connected provider.
No connection
Returns an error with the provider to connect.
A route returns the first target that succeeds. A later target runs only after an eligible provider failure, before the response has started — never to get around a budget or a model policy.
Your accounts. Your keys. Providers bill you directly.
Read the connection guideProviders you can connect
33 kinds today, plus any server that speaks the OpenAI API — self-hosted vLLM, Ollama, LiteLLM, or an internal model behind your own gateway. A connection is one credential for one kind in one environment, and an organization can hold several of the same kind: separate accounts for staging and production, or two Bedrock regions behind one route.
This list is the registry the gateway and the console's connect flow both read, so what is named here is what the picker offers. An operator can register additional kinds for an enterprise agreement.
OpenAI
GPT models, images, speech and transcription on your OpenAI account.
Anthropic
Claude models on your Anthropic account.
Google AI (Gemini)
Gemini models through the Gemini API's OpenAI-compatible endpoint.
Azure OpenAI
Your Azure OpenAI resource and its deployments.
AWS Bedrock
Models on your AWS account, using a Bedrock API key.
Google Vertex AI
Gemini and Claude models on your Google Cloud project, with a service-account key.
Mistral
Mistral models on your Mistral account.
Cohere
Command models through Cohere's OpenAI-compatible endpoint.
Groq
Open-weight models served on Groq hardware, on your Groq account.
Together AI
Open-weight models on your Together account.
Fireworks AI
Open-weight models on your Fireworks account.
DeepSeek
DeepSeek models on your DeepSeek account.
OpenRouter
Any model OpenRouter lists, on your OpenRouter account.
Runway
Runway Gen-4.5 and the video models Runway serves, on your Runway API account.
KIE
Claude, GPT, Grok, Gemini and Kimi chat, and Nano Banana, GPT Image, Seedream and Seedance media, on your KIE account.
LaoZhang API
GPT, Gemini, Grok, DeepSeek, Kimi, GLM and MiniMax chat, plus image models and Seedance video, on your LaoZhang account.
TypeSafe
Jev structured decisions (choice, score, yes/no) on your TypeSafe account.
Cloudflare Workers AI
Clef decisions (yes or no, choice, score) on your Cloudflare account.
Inception
Mercury Decide decisions (yes or no, choice, score) on your Inception account.
Liquid AI
d1 decisions (yes or no, choice, score) on your Liquid AI account.
Upstage
Solar Decide decisions (yes or no, choice, score) on your Upstage account.
Moonshot AI (Kimi)
Kimi models on your Moonshot AI account.
Z.ai (GLM)
GLM models on your Z.ai account.
MiniMax
MiniMax M-series models on your MiniMax account.
Alibaba Cloud Model Studio
Qwen models on your Alibaba Cloud Model Studio account.
Cerebras
Open-weight models served on Cerebras hardware, on your Cerebras account.
Perplexity
Sonar search-grounded models on your Perplexity account.
ElevenLabs
ElevenLabs voices and Scribe transcription on your ElevenLabs account.
fal.ai
Image and video models from many vendors on your fal.ai account.
Replicate
Official image and video models on your Replicate account.
OpenAI-compatible endpoint
A self-hosted or internal server that speaks the OpenAI API: vLLM, Ollama, LiteLLM, TGI, or your own.
What happens to the credential
Handing a provider key to a third party is the part of BYOK that deserves scrutiny, so this is the whole of it.
Sealed before it is stored
AES-256-GCM under a key that is not in the database, with your organization and that connection bound in — so a sealed value copied to another row or another tenant fails to open rather than authenticating as you.
Write-only, with no reveal
Nothing returns the credential after it is saved. The console shows the last four characters so you can tell two keys apart, and offers replace and revoke. There is no button that prints it, for anyone, including us.
Your applications never carry it
They authenticate to INFRO with a scoped gateway key you can rate-limit, restrict to a project or environment, and revoke on its own. The provider credential stops being copied into every service and every laptop.
Used for your requests, and nothing else
INFRO sends requests to this provider on your behalf using this key, and reads the usage it reports. It never reads your provider account, your billing, or anything else.
Leaving is the same shape as arriving: revoke the connection, change the base URL in your application back to the provider's, and your accounts carry on untouched — they were always yours. Trust Center · Subprocessors
A provider that isn't listed
If it speaks the OpenAI API — self-hosted vLLM, Ollama, LiteLLM, or an internal model behind your own gateway — connect it today as an OpenAI-compatible endpoint: its HTTPS base URL, a key if it needs one, and the models you want to route. Your applications then call them through INFRO like any other connection, on your own account. For a provider with its own API format, tell support@infro.io which one you use.
Provider Program
Run an AI API?Have INFRO integrate it.
Apply in about five minutes. An engineer reviews your API, then we build and verify the integration. Teams connect their own account with you and keep paying you directly.
- A Connect page and a Connect to INFRO buttonFor your docs. A reader lands in setup with your API already chosen.
- The INFRO Verified markDated by the last passing run of our suite against your live API. Never sold.
- A private page for your applicationWhere it stands, what we need next, and a safe way to send test access.
- A channel to INFRO's customersDeprecations, incidents and new models, relayed to the teams on your API.
Docs / Integrations
Use your account with INFRO
Route, budget and trace your requests to our API through INFRO, on a key from your own account with us.
- Step 1: ApplyAbout five minutes, from a company address.
- Step 2: ReviewAn engineer reads every application.
- Step 3: Test accessA test key, through your private page.
- Step 4: Integrate and verifyAdapter, model mapping, tests, a live run.
- Step 5: LiveYour pages go up; you get the button.
- No fee to join
- No revenue share
- No referral payment
- No paid placement
- Never changes routing