Introduction
One API key for 60+ AI models, plus built-in RAG, usage limits, team roles and tools.
Tarqa AI is a gateway that sits between your app and AI models. You send every request to one endpoint with one API key, and Tarqa sends it on to Claude, GPT, Gemini, Llama, DeepSeek, or a model running on your own GPU.
Your app
any language
POST /v1/chat/completions →
Tarqa gateway
route · limits · RAG · tools
routes to →
ClaudeGPTGeminiLlamaDeepSeekYour GPU
To switch models you change one field (model). The rest of your code stays the same. The API is OpenAI-compatible, so the official OpenAI SDKs work by just changing the base URL.
Your base URL
Every request in these docs goes to https://api.tarqaai.com/api.
What you get
| Feature | What it does for you |
|---|---|
| 60+ models, one endpoint | Claude Opus 4.6, GPT-4.1, o3, Gemini 2.5 Pro, Llama 4 Maverick, DeepSeek-R1 and more, all called the same way. |
| $0 token markup | You pay plan pricing; Tarqa adds no per-token markup on top. |
| Built-in RAG | Upload documents or web pages and get answers grounded in your own data. |
| Usage limits | Plan-based request and token limits, checked before any call reaches a provider. |
| Teams & roles | 5 roles (Owner, Admin, Developer, Member, Viewer) and SAML SSO with Okta, Azure AD or Google Workspace. |
| Tools | Tool / function calling, plus MCP connectors that Tarqa runs for you. |
| Your own GPUs | Serve local models through a GPU tunnel and call them through the same API. |
| 99.9% uptime SLA | Production-grade availability. |
Start here
Quick StartGet a key and make your first call in about 5 minutes.Chat APISend messages, stream replies, and call tools.All endpointsEvery endpoint at a glance.RAGAnswer questions from your own documents.MCP ConnectorsGive the model tools that run automatically.Code examplesCopy-paste integrations for common use cases.
FAQ
Do I need a new SDK?
No. Tarqa is OpenAI-compatible, so you can point the official OpenAI SDK at your Tarqa base URL. See SDKs & Libraries.
How do I pick a model?
Set the
model field in your request. The full list is in your dashboard under Models, or in Chat API & Models.What happens when I hit my plan limit?
The request is refused with HTTP
429 before it reaches any provider, so you're never charged for it. See Plan Limits.