D1V PAI with CC-Switch and AI CLI clients

D1V PAI exposes an OpenAI-compatible endpoint. The recommended setup depends on the client protocol:

• Codex + CC-Switch: supported directly with the Responses API. This is the simplest path.
• Claude Code + CC-Switch: supported through CC-Switch local routing, which converts Anthropic Messages requests to OpenAI Responses requests. Use CC Switch 3.17.0 or newer.
• CC-Switch CLI/TUI: supported for terminal, SSH, and automation workflows.
• OpenAI-compatible clients: AionUI, OpenCode, Grok Build, WorkBuddy, OpenClaw, and Hermes can use the same Base URL and key when configured for an OpenAI-compatible provider.

Do not point Claude Code directly at D1V_PAI_BASE_URL as an Anthropic endpoint. Claude Code sends /v1/messages, while D1V PAI is provisioned as an OpenAI-compatible gateway. Use the CC-Switch routing instructions below.

1. Sign in and open Dashboard.
2. Wait until the workspace status is ready.
3. Click Create Project Environment in the Dashboard toolbar.
4. Enter a name for the environment and keep D1V PAI selected. Other integrations are optional.
5. Click Create project environment.
6. When the environment is ready, copy these two values:
   • D1V_PAI_BASE_URL: the OpenAI-compatible API URL, normally https://pai.d1v.ai/v1.
   • D1V_PAI_API_KEY: a secret API key scoped to this project environment.

   The values are displayed in full after creation and remain manageable in the project. Use the project key for model calls; do not use or request a platform administration key.

Keep the values in environment variables while testing:

export D1V_PAI_BASE_URL='https://pai.d1v.ai/v1'
export D1V_PAI_API_KEY='<paste-project-key>'

curl "$D1V_PAI_BASE_URL/models" \
  -H "Authorization: Bearer $D1V_PAI_API_KEY"

A successful response contains model IDs under data. Choose a model from this response instead of copying a model name from an old screenshot or third-party tutorial.

Install Node.js first, then install and initialize the CLIs you intend to use:

node --version
npm --version
npm install -g @openai/codex@latest
npm install -g @anthropic-ai/claude-code@latest
codex --help
claude --help

Running the help commands once creates the local configuration directories that CC-Switch manages. If you only need Codex, Claude Code is optional.

Download the current desktop release from CC-Switch Releases. On macOS you can also use:

brew tap farion1231/ccswitch
brew install --cask cc-switch

Open CC-Switch after installation. This guide uses a Custom provider because the built-in PackyCode preset contains PackyAPI endpoints, not D1V PAI endpoints.

1. Open the Codex tab and click + to add an application-specific provider.
2. Keep the Custom preset and name it D1V PAI.
3. Set API Key to the value of D1V_PAI_API_KEY.
4. Set API request URL / Base URL to the exact value of D1V_PAI_BASE_URL (https://pai.d1v.ai/v1).
5. Set Upstream/API format to native Responses.
6. Fetch models from the endpoint or enter a model ID returned by /models as the default model.
7. Save the provider, click Enable, open a new terminal, and run codex.

   If you prefer to inspect the generated configuration, it is equivalent to:

model_provider = "d1v_pai"
model = "<model-id-from-v1-models>"
disable_response_storage = true

[model_providers.d1v_pai]
name = "D1V PAI"
base_url = "https://pai.d1v.ai/v1"
wire_api = "responses"
requires_openai_auth = true

And ~/.codex/auth.json contains:

{
  "OPENAI_API_KEY": "<D1V_PAI_API_KEY>"
}

Claude Code requires protocol conversion. Use CC Switch 3.17.0 or newer:

1. Open the Claude Code tab, click +, and keep the Custom preset.
2. Name the provider D1V PAI and paste D1V_PAI_API_KEY into API Key.
3. Set Request URL to https://pai.d1v.ai. This is the service root derived by removing the trailing /v1 from D1V_PAI_BASE_URL.
4. Expand Advanced options and select OpenAI Responses API (routing required).
5. Keep the authentication field on ANTHROPIC_AUTH_TOKEN (Authorization), so the route sends Authorization: Bearer <key> upstream.
6. Map at least the default fallback model to a model ID returned by D1V PAI /models. You may map Sonnet/Opus to the same main model and Haiku to a smaller model.
7. Open Settings → Routing → Local routing, enable the routing service, then enable routing for Claude Code.
8. Enable the D1V PAI provider, close any running Claude Code process, open a new terminal, and run claude.

   The local route normally listens on 127.0.0.1:15721. CC-Switch keeps the real project key in its provider store and writes only the local route into Claude Code live configuration.

CC-Switch CLI is useful on servers, over SSH, and in automation:

curl -fsSL https://github.com/SaladDay/cc-switch-cli/releases/latest/download/install.sh | bash
cc-switch env tools
cc-switch env check
cc-switch --app codex

In the TUI, add a custom Codex provider using the same Base URL, API Key, Responses format, and model ID described above. For Claude Code, run cc-switch --app claude, configure the custom Responses provider, and enable local routing before testing.

On macOS or Linux, enable the daemon-managed Claude route with:

cc-switch --app claude proxy enable
cc-switch proxy show

On Windows, run the route in the foreground because daemon-managed proxy sessions require Unix domain sockets:

cc-switch proxy serve --takeover claude

Useful checks:

cc-switch --app codex provider current
cc-switch --app codex provider list
cc-switch --app codex env check
cc-switch --app claude env check

For clients with a custom OpenAI-compatible provider, use the same four values:

• Provider/API format: OpenAI compatible
• Base URL: D1V_PAI_BASE_URL
• API Key: D1V_PAI_API_KEY
• Model: an ID returned by GET /models

This applies directly to AionUI, OpenCode, Grok Build, and WorkBuddy. OpenClaw and Hermes are also good candidates when configured through their CC-Switch tabs with an OpenAI-compatible provider.

We do not make Claude Desktop part of the primary setup because it requires additional local routing and model-role mapping. We also keep Claude cache optimization proxy out of this guide: it is an optional performance layer, not required to obtain or validate D1V PAI access.

• 401/403: confirm you copied D1V_PAI_API_KEY, not a D1V account API key or platform administration key.
• 404 or wrong endpoint: Codex and OpenAI-compatible clients use the /v1 Base URL. Claude Code must use CC-Switch routing and the OpenAI Responses format.
• Model not found: refresh GET /models with the same project key and use an exact returned ID.
• Provider switch did not take effect: initialize the target CLI once, enable the provider again, and start a new terminal process.
• Environment overrides configuration: run cc-switch --app codex env check or cc-switch --app claude env check and remove conflicting OPENAI_API_KEY, ANTHROPIC_API_KEY, or Base URL variables.

Never commit the project key to Git, paste it into screenshots, or place it in client-side browser code. Create a separate project environment when you need separate usage tracking or revocation boundaries.

• CC-Switch releases
• CC-Switch provider manual
• PackyAPI CC-Switch common steps
• PackyAPI CC-Switch CLI guide

CC-Switch changes quickly. Field names may move between releases; preserve the protocol choices and values described here even if the form layout changes.