Client integrations¶
Your developers, analysts and tools already call AI models in their own way: an SDK in a service, an agent framework, an assistant in the editor, a chat interface, an automation. Point each one at the gateway, with a Vidai key, and every call is checked against your policies, charged to the right team and recorded in Request logs, without changing how people work.
Two settings, always: the base address (https://<your-gateway>/v1 for OpenAI-format clients,
https://<your-gateway> for Anthropic and Gemini) and a Vidai key (from Keys › Issue a key). The gateway
translates between the OpenAI, Anthropic and Gemini formats, so a client written for one provider can call another's
models.
flowchart LR
A[Your client<br/>SDK, framework, tool] -- base address + Vidai key --> G[Gateway]
G -- policies, guardrails, routing --> P[OpenAI · Anthropic · Gemini<br/>Azure · Vertex AI · Bedrock]
G -. every call .-> R[Request logs]
Start¶
Quickstart with curl · Choosing an SDK · Model names · Endpoints at a glance · Headers and errors
By group¶
When a client can't be pointed at the gateway¶
Some products call models from their own servers with their own keys, and offer no setting for an address: Zapier's and Make's ready-made OpenAI apps, the built-in agents of some cloud platforms, and several editors' own completions. Their pages say so plainly and show what to use instead, usually the tool's plain HTTP step or a client that can. Calls that don't pass through the gateway aren't governed or recorded by Vidai.
Every page has the same shape¶
What it is · the two settings · a minimal call · streaming · tool calls · reading what the gateway did · errors · model names · what goes through the gateway · a one-minute check · good to know.