Google Gemini

Setup

export GEMINI_API_KEY="AIza..."
# or
export GOOGLE_API_KEY="AIza..."

No extra install — included with pip install llmgate.

---

Model prefix

Models starting with gemini- are routed to Gemini automatically.

completion("gemini-2.5-flash-lite", messages)
completion("gemini-2.0-flash", messages)
completion("gemini-2.5-pro", messages)

---

Vision

Gemini does not accept image URLs natively. llmgate fetches URL images client-side using httpx and sends them as inline bytes — this is transparent to you:

completion("gemini-2.0-flash", messages=[{
    "role": "user",
    "content": [
        {"type": "text", "text": "Describe this chart."},
        {"type": "image_url", "image_url": {"url": "https://..."}},  # fetched automatically
    ],
}])

---

Structured outputs

Uses Gemini's native response_schema + application/json — very reliable:

movie = parse("gemini-2.5-flash-lite", messages, response_format=Movie)

---

Notes

• Uses the google-genai SDK (not the older google-generativeai).
• Conversation roles alternate user / model. llmgate converts assistant → model automatically.
• Tool results are sent as function_response parts in a user turn — handled automatically.