Depaza in your IDE.

Claude Code

Environment variables (persist in ~/.claude/settings.json)

1. 1 Install once: npm install -g @anthropic-ai/claude-code (Node 18+).
2. 2 Point it at Depaza's Anthropic-compatible API: export ANTHROPIC_BASE_URL=https://depaza.com — note there is NO /v1, Claude Code appends it.
3. 3 Authenticate: export ANTHROPIC_AUTH_TOKEN=your dpz_live_ key (Depaza also accepts it as ANTHROPIC_API_KEY).
4. 4 Optional — pin a model: export ANTHROPIC_MODEL=core (or max / advanced). Left unset, Claude model names auto-map: anything with "haiku" → lite, everything else → max.
5. 5 Run claude in your project. To make it stick, drop the same values into the env block of ~/.claude/settings.json.

Claude Code uses the Anthropic-compatible endpoint (base https://depaza.com, no /v1) — different from the OpenAI base URL the other editors use. Output is capped at 8192 tokens per turn.

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://depaza.com",
    "ANTHROPIC_AUTH_TOKEN": "dpz_live_your_key",
    "ANTHROPIC_MODEL": "core"
  }
}

Codex CLI

~/.codex/config.toml

1. 1 Install: npm install -g @openai/codex (or brew install codex). Don't run bare codex yet — it opens OpenAI's login.
2. 2 In ~/.codex/config.toml add a [model_providers.depaza] block (see right): base_url https://depaza.com/v1, env_key DEPAZA_API_KEY, wire_api "chat", requires_openai_auth false.
3. 3 Set model_provider = "depaza" and model = "core" (or max / advanced).
4. 4 Export your key so it matches env_key: export DEPAZA_API_KEY=your dpz_live_ key.
5. 5 Run codex — it routes through Depaza with no OpenAI account.

Two fields matter: wire_api = "chat" (Depaza is a Chat Completions API, not the Responses API) and requires_openai_auth = false. The model must be a Depaza id — there is no gpt-4o alias.

model_provider = "depaza"
model = "core"                 # or max, advanced

[model_providers.depaza]
name = "Depaza"
base_url = "https://depaza.com/v1"
env_key = "DEPAZA_API_KEY"
wire_api = "chat"
requires_openai_auth = false

Cursor

Settings → Models → Override OpenAI Base URL (UI only — no config file)

1. 1 Open Settings (Cmd/Ctrl+,) → Models, and turn OFF the built-in models — the override is all-or-nothing.
2. 2 Under the OpenAI provider, click + Add Model and add core, max and advanced (exact names, no vendor prefix).
3. 3 Enable "Override OpenAI Base URL" and enter https://depaza.com/v1 (include /v1, no trailing slash).
4. 4 Paste your dpz_live_ key in the OpenAI API Key field, then click Verify.
5. 5 Open a chat, pick core / max / advanced from the model dropdown and send a prompt.

Cursor Tab, Composer and the Agent use Cursor's own models — the custom endpoint powers Chat and inline edit.

Cline

Cline panel → ⚙ Settings → API Configuration

1. 1 Open the Cline panel in VS Code, then its ⚙ Settings.
2. 2 Set API Provider to "OpenAI Compatible" (not the plain "OpenAI" preset).
3. 3 Base URL: https://depaza.com/v1 · API Key: your dpz_live_ key.
4. 4 Model ID: type core (or max / advanced) exactly — there's no dropdown for custom endpoints.
5. 5 Under Model Configuration, enable function calling so Plan/Act tasks can use tools, then save.

Cline is a full agent and leans on tool-calling — every request goes to Depaza; there's no autocomplete that bypasses it.

Continue

~/.continue/config.yaml — VS Code & JetBrains

1. 1 Command Palette → "Continue: Open configuration file" (config.yaml).
2. 2 Add each model under models: with provider: openai, apiBase: https://depaza.com/v1, your apiKey and model: core / max / advanced.
3. 3 Give each a name and roles: [chat, edit, apply]; add capabilities: [tool_use, image_input] for Agent mode.
4. 4 Save (Continue hot-reloads) and pick "Depaza Core" in the model selector.

Keep Tab / autocomplete on a dedicated fill-in-the-middle model — the chat models aren't built for it.

models:
  - name: Depaza Core
    provider: openai
    model: core
    apiBase: https://depaza.com/v1
    apiKey: dpz_live_your_key
    roles: [chat, edit, apply]
    capabilities: [tool_use, image_input]
  # add max and advanced the same way

Zed

Agent Panel → settings → LLM Providers (or ~/.config/zed/settings.json)

1. 1 Agent Panel → open settings ("agent: open settings") → LLM Providers → + Add Provider (OpenAI-compatible).
2. 2 Provider name: depaza · API URL: https://depaza.com/v1 · API key: your dpz_live_ key.
3. 3 Add models core, max and advanced (set a context window, e.g. 128000).
4. 4 Or edit settings.json directly (right) — the key goes in the OS keychain or the DEPAZA_API_KEY env var, not the file.

Tab (Edit Predictions) is a separate subsystem and stays on its own model.

{
  "language_models": {
    "openai_compatible": {
      "depaza": {
        "api_url": "https://depaza.com/v1",
        "available_models": [
          { "name": "core",     "display_name": "Depaza Core",     "max_tokens": 128000 },
          { "name": "max",      "display_name": "Depaza Max",      "max_tokens": 128000 },
          { "name": "advanced", "display_name": "Depaza Advanced", "max_tokens": 128000 }
        ]
      }
    }
  }
}
// key: set the DEPAZA_API_KEY env var, or enter it in the provider UI

Windsurf

Via the Cline or Roo Code extension (native BYOK is Anthropic-only)

1. 1 Windsurf's native Provider API Keys are Anthropic-only and have no base-URL field — so install the Cline or Roo Code extension from the marketplace.
2. 2 In the extension's settings, set API Provider to "OpenAI Compatible".
3. 3 Base URL: https://depaza.com/v1 · API Key: your dpz_live_ key.
4. 4 Model ID: core (or max / advanced), then send a test message.

The extension runs as its own panel; Windsurf's Cascade agent and Tab stay on first-party models.