One API key, all models
Unified billing, unified observability. Switch models with a single line change — no new SDKs, no new credentials.
Built-in failovers
Automatic fallback during provider outages. Define failover chains per model to keep your app running through any disruption.
No markup, list price
Zero platform fee on model usage. You pay the exact same price as going direct to each provider — we make money on the platform, not the gateway.
Drop-in compatible
OpenAI-compatible API. Works with any SDK that supports custom base URLs.
Chat completions across providers
Text and tool-use models from three vendors today. Add anything else through OpenAI-compatible BYOK.
Chat, completion, reasoning. 19 text models from OpenAI, Anthropic, and Google.
Code generation, completion, and refactoring with reasoning-tuned models.
Bring your own image-model endpoint via the openai-compatible provider type.
Local embeddings via HuggingFace TEI power the knowledge base; bring your own for app code via BYOK.
Built for production
Everything you need to run AI in production, without the complexity.
Usage analytics
Track token usage, costs, latency, and error rates per model, per team, per project.
Rate limiting
Set per-team or per-project rate limits and spending caps. No surprise bills.
Automatic retries
Retry failed requests with exponential backoff. Configure fallback models for resilience.
Zero data retention
We never store prompts or completions. Requests are proxied and forgotten.
Streaming support
Full SSE streaming for all text and code models. Same stream format across all providers.
Semantic caching
Cache identical or semantically similar requests to reduce costs and latency.
No markup, no platform fee
The AI Gateway is free to use. You pay the upstream provider's list price — nothing more. We make money from the proxifai platform, not the gateway.
Frequently asked questions
How is the gateway free?
The gateway itself has no platform fee. You pay the exact list price from each provider. We monetize through the proxifai platform (agent compute, collaboration, etc.), and offer the gateway as infrastructure for developers using any part of our ecosystem.
Do you impose rate limits?
We don't impose our own rate limits on top of provider limits. You get the full rate limits of your upstream provider tier. You can optionally configure spending caps and per-team limits for cost control.
Do you store my prompts or completions?
No. Zero data retention by default. Requests are proxied to the provider and immediately discarded. We only store aggregate metadata (token counts, latency, error codes) for analytics — never prompt content.
Can I bring my own API keys?
Yes. You can use our pooled access (pay-as-you-go via proxifai billing) or bring your own provider API keys. BYOK mode routes through the gateway for unified analytics and failovers while billing directly to your provider account.
How many models are supported?
19 models ship out of the box across three providers: Anthropic (Opus 4.6, Sonnet 4.6/4.5, Haiku 4.5), OpenAI (GPT-4.1 family, GPT-4o family, o1/o3/o4 reasoning), and Google (Gemini 2.5 and 2.0). Anything OpenAI-compatible can be added at runtime via BYOK with a custom base URL.
Start routing requests
Get an API key and start using any model in under a minute.