Coding agents

Every AI coding tool that speaks the OpenAI or Anthropic API works with Depaza. Point it at our endpoint, give it your API key, and pick a model.

Set up once

Enter your key and pick a model. Every config block across these pages fills itself in with your real values. Get a key from the API Console.

Depaza API key

Model

Your key stays in this browser (local storage) and is sent only to depaza.eu. Test connection sends one real, billed or metered chat request. Don’t do this on a shared machine.

Which endpoint

Two endpoints, depending on what the tool speaks. Most tools want OpenAI. Claude Code wants Anthropic.

	Endpoint	Auth
OpenAI-compatible	`https://depaza.eu/v1`	`Authorization: Bearer dpz_live_…`
Anthropic-compatible	`https://depaza.eu`	`ANTHROPIC_AUTH_TOKEN` / `x-api-key`

Claude Code CLI and VS Code extension Cursor Model override in settings opencode Custom provider block GitHub Copilot Custom endpoint in VS Code Cline VS Code agent Roo Code VS Code agent Kilo Code VS Code agent Zed Editor with built-in agent aider Terminal pair programmer goose Block’s terminal agent Continue VS Code & JetBrains Codex CLI No OpenAI login Windsurf Via Cline or Roo Vercel AI SDK Build your own agent

Choosing a model

id
`core`	Default. Everyday agent work.
`coder`	Tool-heavy coding loops.
`max`	Harder multi-file changes and deeper reasoning.
`advanced`	Hardest problems. 3× tokens, 16384 max output.
`lite`	Fast and cheap. Haiku stand-in.

IDs are lowercase and have no vendor prefix. Tools that ask for a context window by hand should use 128000, and max output 8192 (16384 for advanced).

curl

curl {{BASE_URL}}/chat/completions \
  -H "Authorization: Bearer {{KEY}}" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "{{MODEL}}",
    "messages": [{"role": "user", "content": "Say hello"}]
  }'