Skip to content

SETUP.md

Latest commit

 

History

History
197 lines (175 loc) · 7.34 KB

SETUP.md

File metadata and controls

197 lines (175 loc) · 7.34 KB
Raw

DroidProxy Setup

1. Launch & Authenticate

  1. Open DroidProxy from your Applications folder
  2. Click the menu bar icon and select "Open Settings"
  3. Click "Connect" next to Claude Code, Codex, Gemini, Kimi, or GitHub Copilot (browser/device-code OAuth), Junie (API key), or Grok (device-code browser login)

GitHub Copilot

  1. Connect GitHub Copilot in Settings and complete the device-code sign-in.
  2. After the local Copilot gateway starts, click Refresh Models.
  3. Choose up to three models available to your Copilot subscription.
  4. Click Apply / Re-apply under Factory custom models.

Only these selected Copilot models are written into ~/.factory/settings.json. The gateway runs locally on 127.0.0.1:8319; no Copilot credential is added to Factory settings. GitHub Copilot support requires Node.js 20 or later so DroidProxy can run the maintained local gateway. Node.js 22.13 or later additionally enables the gateway's local usage storage.

For Grok 4.6 (SuperGrok / X Premium+ OAuth → api.x.ai):

  1. Connect Grok in Settings (device-code browser login)
  2. Click Apply / Re-apply under Factory models (registers custom:droidproxy:grok-4.6)
  3. In Droid, /model → DroidProxy: Grok 4.6

Optional Grok 4.6 Fast Mode (Settings → Grok): rewrites grok-4.6 → grok-4.6-fast through the Cursor API. Enable Beta → Cursor and add a Cursor API key first; api.x.ai does not offer grok-4.6-fast. Default off.

Note: some SuperGrok tiers return HTTP 403 on the OAuth API surface even after a successful login. Fallback is an XAI_API_KEY via Factory BYOK.

Grok Imagine (image generation)

Chat and image generation share the same Grok OAuth session, but Droid will not call the image API by itself. Install the skill from this repo:

mkdir -p ~/.factory/skills
cp -R skills/grok-imagine ~/.factory/skills/

See skills/grok-imagine/SKILL.md. With DroidProxy running and Grok connected, the skill posts to http://localhost:8317/v1/images/generations (grok-imagine-image-2.0); the proxy injects your subscription bearer and forwards to api.x.ai.

2. Configure Factory

Open ~/.factory/settings.json and add the following to the customModels array:

"customModels": [
    {
      "model": "claude-fable-5-1",
      "id": "custom:droidproxy:fable-5-1",
      "index": 0,
      "baseUrl": "http://localhost:8317",
      "apiKey": "dummy-not-used",
      "displayName": "DroidProxy: Fable 5.1",
      "maxOutputTokens": 128000,
      "noImageSupport": false,
      "provider": "anthropic"
    },
    {
      "model": "claude-fable-5",
      "id": "custom:droidproxy:fable-5",
      "index": 1,
      "baseUrl": "http://localhost:8317",
      "apiKey": "dummy-not-used",
      "displayName": "DroidProxy: Fable 5",
      "maxOutputTokens": 128000,
      "noImageSupport": false,
      "provider": "anthropic"
    },
    {
      "model": "claude-opus-5",
      "id": "custom:droidproxy:opus-5",
      "index": 2,
      "baseUrl": "http://localhost:8317",
      "apiKey": "dummy-not-used",
      "displayName": "DroidProxy: Opus 5",
      "maxOutputTokens": 128000,
      "noImageSupport": false,
      "provider": "anthropic"
    },
    {
      "model": "claude-sonnet-4-6",
      "id": "custom:droidproxy:sonnet-4-6",
      "index": 3,
      "baseUrl": "http://localhost:8317",
      "apiKey": "dummy-not-used",
      "displayName": "DroidProxy: Sonnet 4.6",
      "maxOutputTokens": 64000,
      "noImageSupport": false,
      "provider": "anthropic"
    },
    {
      "model": "gpt-5.4",
      "id": "custom:droidproxy:gpt-5.4",
      "index": 4,
      "baseUrl": "http://localhost:8317/v1",
      "apiKey": "dummy-not-used",
      "displayName": "DroidProxy: GPT 5.4",
      "maxOutputTokens": 128000,
      "noImageSupport": false,
      "provider": "openai"
    },
    {
      "model": "gpt-5.5",
      "id": "custom:droidproxy:gpt-5.5",
      "index": 5,
      "baseUrl": "http://localhost:8317/v1",
      "apiKey": "dummy-not-used",
      "displayName": "DroidProxy: GPT 5.5",
      "maxOutputTokens": 128000,
      "noImageSupport": false,
      "provider": "openai"
    },
    {
      "model": "gpt-5.6-terra",
      "id": "custom:droidproxy:gpt-5.6-terra",
      "index": 6,
      "baseUrl": "http://localhost:8317/v1",
      "apiKey": "dummy-not-used",
      "displayName": "DroidProxy: GPT 5.6 Terra",
      "maxOutputTokens": 128000,
      "noImageSupport": false,
      "provider": "openai"
    },
    {
      "model": "gpt-5.6-sol",
      "id": "custom:droidproxy:gpt-5.6-sol",
      "index": 7,
      "baseUrl": "http://localhost:8317/v1",
      "apiKey": "dummy-not-used",
      "displayName": "DroidProxy: GPT 5.6 Sol",
      "maxOutputTokens": 128000,
      "noImageSupport": false,
      "provider": "openai"
    },
    {
      "model": "gemini-3.1-pro-preview",
      "id": "custom:droidproxy:gemini-3.1-pro",
      "index": 8,
      "baseUrl": "http://localhost:8317",
      "apiKey": "dummy-not-used",
      "displayName": "DroidProxy: Gemini 3.1 Pro",
      "maxOutputTokens": 65536,
      "noImageSupport": false,
      "provider": "google"
    },
    {
      "model": "gemini-3-flash-preview",
      "id": "custom:droidproxy:gemini-3-flash",
      "index": 9,
      "baseUrl": "http://localhost:8317",
      "apiKey": "dummy-not-used",
      "displayName": "DroidProxy: Gemini 3 Flash",
      "maxOutputTokens": 65536,
      "noImageSupport": false,
      "provider": "google"
    },
    {
      "model": "kimi-k3",
      "id": "custom:droidproxy:kimi-k3",
      "index": 10,
      "baseUrl": "http://localhost:8317/v1",
      "apiKey": "dummy-not-used",
      "displayName": "DroidProxy: Kimi K3",
      "maxOutputTokens": 65536,
      "noImageSupport": false,
      "provider": "openai",
      "enableThinking": true,
      "supportedReasoningEfforts": ["max"],
      "defaultReasoningEffort": "max",
      "reasoningEffort": "max"
    }
]

Use the standard Claude, Codex, Gemini, and Kimi model aliases in the model field. Claude and Gemini entries use http://localhost:8317 (with provider: "anthropic" and provider: "google" respectively); GPT/Codex and Kimi entries use provider: "openai" with http://localhost:8317/v1. Reasoning effort is chosen per session from Droid CLI's model selector. DroidProxy preserves that selection, and CLIProxyAPI translates it to the provider-native request format.

3. Choose Reasoning Effort

Reasoning effort is selected per session in Droid CLI's model picker. DroidProxy registers each model with its native reasoning levels, preserves the selected value, and lets CLIProxyAPI translate it to the upstream provider format. Supported levels per model:

  • Fable 5.1: low, medium, high, xhigh, or max
  • Fable 5: low, medium, high, xhigh, or max
  • Opus 5: low, medium, high, xhigh, or max
  • Sonnet 4.6: low, medium, high, or max
  • GPT 5.4: low, medium, high, or xhigh
  • GPT 5.5: low, medium, high, or xhigh
  • GPT 5.6 Terra: none, low, medium, high, xhigh, or max
  • GPT 5.6 Sol: dynamic, low, medium, high, xhigh, or max
  • Gemini 3.1 Pro: low, medium, or high
  • Gemini 3 Flash: minimal, low, medium, or high
  • Kimi K3: max (the only currently supported effort)
  • Kimi K2.6: high

4. Enable Thinking Output

  1. Start Factory
  2. Run /settings
  3. Set Show thinking in main view: On