How Antigravity Add Model works
Three tracks, one goal: your providers inside Antigravity.
Three tracks
Desktop, IDE, or gateway
Each track integrates differently — pick yours, then follow its guide.
Standalone desktop app
Desktop
The addon appends its Settings controls and preserves the installed vendor runtime. Custom models route through validated hooks —no proxy required for custom models.
VS Code-based
Antigravity IDE
The IDE uses a settings adapter and an owned local proxy. The proxy starts with a managed lifecycle and uses its own data directory.
Optional
Gateway
The gateway is a separate Node process for centralized routing, browser management, request history, and cost estimates.
Does this modify the Antigravity app itself?
On the desktop, the patch is additive — it inserts validated hooks and preserves the vendor runtime, with version-specific backups and rollback on failure. On the IDE, it writes a reversible setting and starts an independent proxy. The original app is never replaced.
Source: /docs/compatibility/
Under the hood
What the installer actually does
The same four moves as the quickstart, with the mechanism behind each.
Check
node scripts/deploy.mjs --check reads your installation and validates three integration hooks without changing files.
Patch
The installer adds its custom-model modules under dist/modelPatch, preserves vendor preload, IPC handlers, and certificate handling, and verifies the candidate before replacement.
Configure
Model entries live in a JSON file at ~/.gemini/antigravity/custom_models.json. Provider presets fill in defaults; credentials are encrypted at rest.
Launch
The proxy intercepts the model-list endpoint, merges custom entries into the catalog, and routes generation requests to your configured provider.
Comparison
Stock Antigravity vs the patched app
What the patch adds on top of the installed app — seven rows from the repo's capability table and feature comparison.
| What you get | Stock Antigravity | Antigravity Add Model |
|---|---|---|
| Custom providers (BYOK) | No | Yes |
| Local models (Ollama, LM Studio, vLLM) | No | Yes |
| 40+ provider presets | No | Yes |
| Model catalog discovery | No | Yes |
| Fallback chains for rate limits | No | Yes |
| Web dashboard gateway (optional) | No | Yes |
| License | Proprietary | Apache-2.0 patch |
Rows trace to the repo's README capability table and the source-basedfeature comparison — read it for implementation boundaries and scope.
Ready to add your first model?
Two commands check your version, then your provider appears in your dropdown.