Mythical Pro Agents

Gemini Integration

GitHub Edit page

Gemini Integration

Oracle supports Gemini in two distinct ways:

  1. Gemini API mode (--engine api) via GEMINI_API_KEY
  2. Gemini web (cookie) mode (--engine browser) via your signed-in Chrome cookies at gemini.google.com (no API key required)

#Usage (API)

  1. Get an API Key: Obtain a key from Google AI Studio.
  2. Set Environment Variable: Export the key as GEMINI_API_KEY.

    3. ``bash export GEMINI_API_KEY="your-google-api-key" ``

  3. Run Oracle: Use the --model (or -m) flag to select Gemini.

    4. ``bash oracle --engine api --model gemini --prompt "Explain quantum entanglement" ` Use an explicit current model ID: `bash oracle --engine api --model gemini-3.5-flash --prompt "..." ` Gemini 3.1 Pro is also available; Oracle dispatches it to Google's preview model id: `bash oracle --engine api --model gemini-3.1-pro --prompt "..." ` For the lowest-cost current model: `bash oracle --engine api --model gemini-3.1-flash-lite --prompt "..." ``

#Usage (Gemini web / cookies)

Gemini web mode is a cookie-based client for gemini.google.com. It does not use GEMINI_API_KEY and does not drive ChatGPT.

Prereqs:

  • Chrome installed.
  • Signed into gemini.google.com in the Chrome profile Oracle uses (default: Default profile).

Examples:

# Text run
oracle --engine browser --model gemini-3.5-flash --prompt "Say OK."

# Deep Think browser run (manual-login profile recommended on macOS)
oracle --engine browser --browser-manual-login \
  --model gemini-3-deep-think \
  --prompt "Think carefully, then answer in one paragraph."

# Generate an image (writes an output file)
oracle --engine browser --model gemini-3.1-pro \
  --prompt "a cute robot holding a banana" \
  --generate-image out.jpg --aspect 1:1

# Edit an image (input via --edit-image, output via --output)
oracle --engine browser --model gemini-3.1-pro \
  --prompt "add sunglasses" \
  --edit-image in.png --output out.jpg

Notes:

  • Current explicit IDs are gemini-3.1-flash-lite, gemini-3.5-flash, and gemini-3.1-pro.
  • Legacy gemini-3-pro, gemini-2.5-pro, and gemini-2.5-flash browser names remain accepted and map to current Gemini web models.
  • If your logged-in Gemini account can’t access the requested model, Oracle auto-falls back to Gemini 3.1 Flash-Lite and logs the fallback in verbose mode.
  • Pass --no-gemini-fallback to fail instead when the requested web model is unavailable.
  • Gemini web requests accept response headers up to 64 KiB automatically, including Google's large security and reporting policies; no Node options are required. These requests honor HTTP_PROXY, HTTPS_PROXY, and NO_PROXY (including lowercase equivalents) and Node's TLS trust configuration.
  • This path runs fully in Node/TypeScript (no Python/venv dependency).
  • Local MCP consultations and detached workers use the same Gemini executor as the CLI and preserve the session's saved Gemini options. The explicit model key takes precedence over ChatGPT picker labels.
  • Gemini browser runs also support --remote-host. Upgrade both endpoints, sign into Gemini on the service host, then run oracle --engine browser --model gemini-3.5-flash --remote-host <host:port> --remote-token <token> --prompt "Say OK.". New runs and restarts dispatch the Gemini web executor on that host; Google cookies and browser state stay there. Host manual-login profiles and oracle serve --browser-attach-running are supported. Text, file attachments, YouTube prompts, thoughts, and --no-gemini-fallback are forwarded; image generation/editing still require a local Gemini run.
  • --browser-model-strategy only affects ChatGPT automation; Gemini web always uses the explicit Gemini model ID.
  • gemini-3-deep-think is browser-only for now. --engine api rejects it instead of silently falling back to regular Gemini Pro.
  • Oracle intentionally does not expose generic low / medium / high Gemini aliases. Explicit IDs keep model choice, billing, and thinking-effort configuration distinct.
  • If Chrome cookie extraction fails, the missing-cookie error now includes any cookie-reader warnings plus --browser-manual-login / --browser-inline-cookies-file guidance.

#Implementation details

#Gemini API adapter

  • src/oracle/gemini.ts — adapter using @google/genai that returns a ClientLike.
  • Model IDs: gemini-3.1-flash-lite and gemini-3.5-flash use their stable API IDs; gemini-3.1-pro maps to gemini-3.1-pro-preview; legacy gemini-3-pro maps to gemini-3-pro-preview.
  • Request mapping: OracleRequestBody → Gemini request; web_search_preview maps to Gemini search tooling.
  • Response mapping: Gemini responses → OracleResponse.
  • Streaming: wraps Gemini’s async iterator as ResponseStreamLike.
  • src/oracle/run.ts — selects GEMINI_API_KEY vs OPENAI_API_KEY based on model prefix.
  • src/oracle/config.ts / src/oracle/types.ts — model config + ModelName.
  • src/gemini-web/models.ts — centralizes current private web model headers, legacy aliases, and fallback selection.
  • src/gemini-web/client.ts — talks to gemini.google.com and downloads generated images via authenticated gg-dl redirects.
  • src/gemini-web/http.ts — uses a reusable, Gemini-only Undici dispatcher with a 64 KiB response-header limit. It does not change the global dispatcher or replay failed requests. Embedders with custom dispatcher/proxy/TLS settings can use createGeminiWebDispatcher(options) and pass dispatcher to fetchGeminiWebResource; an explicitly supplied dispatcher owns its header limit. Global dispatcher customizations do not configure this dedicated transport.
  • src/gemini-web/executor.ts — browser-engine executor for Gemini (loads Chrome cookies and runs the web client).

#Testing

  • Unit/regression: pnpm vitest run tests/gemini.test.ts tests/gemini-web
  • Live (API): ORACLE_LIVE_TEST=1 pnpm vitest run tests/live/gemini-live.test.ts
  • Live (Gemini web/cookies): ORACLE_LIVE_TEST=1 pnpm vitest run tests/live/gemini-web-live.test.ts