sksarvesh007/antiglaude

What it does

Antiglaude is a local FastAPI gateway that accepts Anthropic Messages (/v1/messages) and OpenAI Chat Completions (/v1/chat/completions) requests and translates them into Google Cloud Code and Gemini execution formats. It targets Claude Code CLI and OpenAI-compatible SDKs, letting them talk to model IDs such as claude-sonnet-4-6, claude-opus-4-6-thinking, gemini-3-flash and gemini-2.5-pro.

The bridge exists because the two sides disagree on protocol details: upstream tokens expire, Google's protobuf serializers reject JSON Schema keywords such as propertyNames, anyOf and const, and long reasoning turns drop connections. The gateway normalizes schemas, refreshes tokens, emits SSE keep-alives, and rewrites identity and billing headers (overview). It also includes a terminal dashboard for quota, latency and request logs, a /model picker for Claude Code, an optional macOS launchd service, and an OpenAI-compatible endpoint (README).

Quick start

Requirements: Python 3.10+, an existing Google Antigravity / Cloud Code login on the machine, and Claude Code if you want the CLI integration. Install from PyPI or with uv:

pip install antiglaude
# or: uv tool install antiglaude

antiglaude start        # gateway on http://127.0.0.1:8080
antiglaude tui          # live control panel
antiglaude setup-claude # write ~/.claude/settings.json and /model options

After setup-claude, start claude from any shell and use /model to pick a backend. The written config sets ANTHROPIC_BASE_URL to the local gateway and supplies a picker manifest (config example). To test the OpenAI side:

curl -X POST http://127.0.0.1:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer test" \
  -d '{"model":"gemini-3-flash","messages":[{"role":"user","content":"Ping"}]}'

Source checkouts use make start, make setup-claude and make daemon-install (quick start). The README does not document which local state files hold the upstream credentials, so the first-run authentication path beyond make status is not fully specified.

Constraints

• Binding: the server listens on loopback only (127.0.0.1).
• Platform: the background daemon is macOS launchd-specific; Windows and Linux users run the gateway in the foreground.
• Credentials: the bridge reads and refreshes OAuth tokens from local application state, so it depends on an existing Antigravity/Cloud Code session rather than an API key you supply.
• Upstream behavior: model IDs are aliases mapped to Google backends, and arbitrary client model names are mapped heuristically to the most capable active upstream model (model directory).
• Quota: the dashboard tracks the documented 1,500 requests/day limit on the upstream plan; actual limits come from the upstream service.
• License: MIT.

Accounts used through the bridge remain subject to Google's and Anthropic's terms. The README describes identity-marker neutralization and local safeguard handling (security notes, auto-mode section); these are compatibility mechanics, not a statement that upstream providers permit the routing.

HotRepo's take

Antiglaude first appears on HotRepo's new-project board on 2026-11-06, with 4 stars gained in the observed two-hour window and 105 stars total. That is a short window and a single snapshot, so it says little about sustained traction. The project is technically concrete: it has a PyPI package, CI, a test harness the README says covers 18 unit tests, and a clear protocol-translation story. Its value depends heavily on the upstream routing remaining viable under provider terms and quota changes, which is the main risk to weigh before adopting it.