Skip to content

studio.md

Latest commit

 

History

History
86 lines (66 loc) · 2.82 KB

studio.md

File metadata and controls

86 lines (66 loc) · 2.82 KB
Raw

OpenAI-Compatible Client Integration

This proxy speaks the OpenAI HTTP API. Any client that supports a custom OpenAI base URL can be pointed at it. Below is one worked example; the same settings work for OpenAI's official Python SDK, openai-cli, ChatBox, NextChat, LobeChat, and others.

Settings (worked example)

Field Value
Provider Custom OpenAI-compatible (or "OpenAI" with a custom base URL)
Base URL http://localhost:4982/openai
API Key Any non-empty string (e.g. test). The proxy does not verify it by default.
Chat model gemini-3-flash / gemini-3-pro (use whatever /openai/v1/models advertises)
Image model gemini-2.5-flash-image / gemini-2.5-pro-image (stable aliases — chat ids won't be routed to the image pipeline)

The Base URL must end with /openai. Most clients append /v1/chat/completions and /v1/images/generations automatically.

Response Conventions

  • Images: only b64_json is returned. Google CDN url is intentionally not exposed, because downstream clients cannot usually fetch lh3.googleusercontent.com/... links.
  • Chat: stream=false only in this version. Streaming is on the roadmap — see README.md §Roadmap.
  • Reference images: send as the request body's image field (data URL array). The proxy forwards them as attachments to Gemini Web.

Concurrency

The proxy runs a single shared GeminiClient per process; there is no request queue. Set your client's image generation concurrency to 1–2 to avoid cookie-refresh storms that surface as 500/401.

Smoke Test

After starting the proxy, verify the health endpoint:

curl -sS http://localhost:4982/health
# {"status":"ok","service":"gemini-webapi-proxy"}

Then trigger a simple chat completion through your client to confirm end-to-end wiring before relying on it for real work.

Using the OpenAI Python SDK

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:4982/openai/v1",
    api_key="not-verified",  # any non-empty string
)

resp = client.chat.completions.create(
    model="gemini-3-flash",
    messages=[{"role": "user", "content": "Reply with exactly: pong"}],
)
print(resp.choices[0].message.content)

Reference-image example

import base64, pathlib
from openai import OpenAI

img_b64 = base64.b64encode(pathlib.Path("ref.png").read_bytes()).decode()
data_url = f"data:image/png;base64,{img_b64}"

client = OpenAI(base_url="http://localhost:4982/openai/v1", api_key="x")
resp = client.images.generate(
    model="gemini-3-flash",
    prompt="a portrait in the same style as the reference",
    n=1,
    size="1024x1024",
    extra_body={"image": [data_url]},
)
import base64 as b
pathlib.Path("out.png").write_bytes(b.b64decode(resp.data[0].b64_json))