AIStudio2API's Build chat models carry the upstream coding assistant's system prompt. The English README says that prompt shapes the answer's style, format and persona. If your application needs control over those behaviors, this is an adoption constraint even when the request format matches your client.
What the proxy connects
AIStudio2API's English README describes a Go service that converts Google AI Studio's web protocol into OpenAI Chat Completions, OpenAI Responses, Anthropic Messages and Gemini GenerateContent interfaces. It also documents streaming, tool calls and agent clients including Claude Code. Those are project claims, rather than authenticated compatibility results from the supplied run.
This review is for developers who already want AI Studio as their upstream. The decision involves more than selecting an API format: the documented setup saves Google account state, and Build chat introduces an upstream prompt that your application does not solely control.
What happened when we ran it
The supplied sandbox run installed AIStudio2API commit a784c51 in 48 seconds and built it in 6 seconds on October 18, 2026. Installation recorded 87 dependencies. The test command failed with exit 1 after 60 seconds; the supplied summary reports 6 passed and 4 failed out of 10.
The log tail names setup failures in internal/app and internal/webui, plus TestUserCacheRootWithoutEnvironment under internal/aistudio. It also includes successful package results for internal/api, internal/camoufoxnative, internal/config, internal/proxydial, internal/requestdb and internal/setup. The summary does not establish ten individual test assertions. The excerpt omits the underlying setup errors and the cache test's assertion details, so it cannot establish their causes.
The recorded sandbox configuration was 3 CPUs, 8g RAM and the golang:1.24-bookworm image label. The label does not establish the actual toolchain or operating-system version. The checkout contained 389 files, approximately 129,002 source lines and 9.6 MB. The scan found one CI workflow, a Dockerfile and a Compose file, with no tests directory; the test log nevertheless shows tests in internal packages. Installed disk footprint and vulnerability results were not measured in the supplied record.
Account setup and access controls
The documented login reads the Google email from AI Studio, authorizes Google Drive and saves the account under the path configured by AISTUDIO_AUTH_STATES. Organizations should check whether that account access and persistence fit their policies before onboarding an account. Treat the saved state as a credential.
For Docker, the README says to set PROXY_API_KEY when the port is reachable from other machines. Opening the console from outside the container requires ADMIN_AUTH_ENABLED=true and a console password; the documented default username is admin. These instructions distinguish access to the proxy API from access to account management. They do not establish a passwordless management default.
The browser boundary depends on the operation
The README names Camoufox as the default backend. With WAA_BACKEND=go, it says no browser is downloaded or launched at runtime. Its Linux and macOS account-setup instructions still use an isolated Camoufox login. Windows also has a documented local Chrome account-import route. Choose an onboarding route separately from the runtime backend.
Source installation includes frontend dependencies, a frontend build and a Go build. The README lists Go 1.26.0+ and Node.js 22.13+ or 24+ for source builds. These documented requirements should not be confused with the sandbox image label or treated as toolchain versions verified by the run.
Build chat limits prompt control
Build chat models include the upstream coding assistant's system prompt. The README explicitly warns that it affects answer style, format and persona. Applications with strict output or identity requirements should evaluate the selected Build chat model against those requirements.
The supplied v0.2.13 release page identifies a784c51, matching the measured checkout. It therefore provides no basis for treating that release as a later fix for the observed test failure. Release packaging and authenticated operation were not tested in the supplied run.
When to consider another proxy
LiteLLM provides a Python SDK and an AI gateway. Consider it when your requirement is direct library integration or a centralized gateway across model providers.
CLIProxyAPI provides compatible API interfaces for multiple CLI accounts. Consider it when you need an endpoint serving several account-backed CLI integrations. Neither alternative was measured in this sandbox run, so this comparison concerns documented scope rather than tested reliability or performance.
For an AIStudio2API trial, first review the account permissions and investigate the recorded test failures. Then check one complete conversation through your intended client, account, model and channel, including a tool call and its continuation. That is a proposed next check, not a result established here.