@s3rezhkaa/opencode-antigravity-auth
v1.7.0
Published
Dignity fork: Google Antigravity OAuth plugin for Opencode with Gemini 3.5 models
Maintainers
Readme
Antigravity + Gemini CLI OAuth Plugin for Opencode
Enable Opencode to authenticate against Antigravity (Google's IDE) via OAuth so you can use Antigravity rate limits and access current models like gemini-3.1-pro, gemini-3.5-flash, claude-sonnet-4-6, and gpt-oss-120b-medium with your Google credentials.
What You Get
- Gemini 3.1 Pro, Gemini 3.5 Flash, Claude Opus/Sonnet 4.6 Thinking, and GPT-OSS 120B via Google OAuth
- Multi-account support — add multiple Google accounts, auto-rotates when rate-limited
- Antigravity quota routing — current defaults avoid Gemini CLI models because individual Gemini CLI access sunsets on 2026-06-18
- Thinking models — extended thinking for Claude and Gemini 3 with configurable budgets
- Google Search grounding — enable web search for Gemini models (auto or always-on)
- Auto-recovery — handles session errors and tool failures automatically
- Plugin compatible — works alongside other OpenCode plugins (oh-my-opencode, dcp, etc.)
[!CAUTION] Using this plugin (and any proxy for Antigravity) violates Google's Terms of Service. A number of users have reported their Google accounts being banned or shadow-banned (restricted access without explicit notification).
By using this plugin, you acknowledge:
- This is an unofficial tool not endorsed by Google
- Your account may be suspended or permanently banned
- You assume all risks associated with using this plugin
Installation
Option A: Let an LLM do it
Paste this into any LLM agent (Claude Code, OpenCode, Cursor, etc.):
Install the opencode-antigravity-auth plugin and add the Antigravity model definitions to ~/.config/opencode/opencode.json by following: https://raw.githubusercontent.com/NoeFabris/opencode-antigravity-auth/dev/README.mdOption B: Manual setup
Add the plugin to
~/.config/opencode/opencode.json:{ "plugin": ["opencode-antigravity-auth@latest"] }Want bleeding-edge features? Use
opencode-antigravity-auth@betainstead.Login with your Google account:
opencode auth loginAdd models — choose one:
- Run
opencode auth login→ Google → OAuth with Google (Antigravity) → select "Configure models in opencode.json" (auto-configures all models) - Or manually copy the full configuration below
- Run
Use it:
opencode run "Hello" --model=google/antigravity-claude-opus-4-6-thinking
Step-by-Step Instructions
Edit the OpenCode configuration file at
~/.config/opencode/opencode.jsonNote: This path works on all platforms. On Windows,
~resolves to your user home directory (e.g.,C:\Users\YourName).Add the plugin to the
pluginarrayAdd the model definitions from the Full models configuration section
Set
providerto"google"and choose a model
Verification
opencode run "Hello" --model=google/antigravity-claude-opus-4-6-thinkingModels
Model Reference
Current Antigravity quota set (from the Antigravity Model Quota view):
| Model ID | Quota row |
|----------|-----------|
| antigravity-gemini-3.1-pro-low | Gemini 3.1 Pro (Low) |
| antigravity-gemini-3.1-pro-high | Gemini 3.1 Pro (High) |
| antigravity-claude-sonnet-4-6 | Claude Sonnet 4.6 |
| antigravity-claude-opus-4-6-thinking | Claude Opus 4.6 (Thinking) |
| antigravity-gpt-oss-120b-medium | GPT-OSS 120B (Medium) |
| antigravity-gemini-3.5-flash-low | Gemini 3.5 Flash (Low) |
| antigravity-gemini-3.5-flash-high | Gemini 3.5 Flash (High) |
| antigravity-gemini-3.5-pro-low | Gemini 3.5 Pro (Low) |
| antigravity-gemini-3.5-pro-high | Gemini 3.5 Pro (High) |
Gemini CLI sunset: Gemini CLI quota models are no longer included in the default model list. Google announced Gemini CLI access for individual/free Google AI Pro/Ultra and Gemini Code Assist users stops on 2026-06-18. Legacy Gemini CLI routing remains in the plugin for backward compatibility only.
Use a current model:
opencode run "Hello" --model=google/antigravity-gemini-3.5-flash-low
opencode run "Hello" --model=google/antigravity-claude-sonnet-4-6For details on compatibility with older variant/tier names, see docs/MODEL-VARIANTS.md.
Add this to your ~/.config/opencode/opencode.json:
{
"$schema": "https://opencode.ai/config.json",
"plugin": [
"opencode-antigravity-auth@latest"
],
"provider": {
"google": {
"models": {
"antigravity-gemini-3.1-pro-low": {
"name": "Gemini 3.1 Pro Low (Antigravity)",
"limit": {
"context": 1048576,
"output": 65535
},
"modalities": {
"input": [
"text",
"image",
"pdf"
],
"output": [
"text"
]
}
},
"antigravity-gemini-3.1-pro-high": {
"name": "Gemini 3.1 Pro High (Antigravity)",
"limit": {
"context": 1048576,
"output": 65535
},
"modalities": {
"input": [
"text",
"image",
"pdf"
],
"output": [
"text"
]
}
},
"antigravity-claude-sonnet-4-6": {
"name": "Claude Sonnet 4.6 (Antigravity)",
"limit": {
"context": 200000,
"output": 64000
},
"modalities": {
"input": [
"text",
"image",
"pdf"
],
"output": [
"text"
]
}
},
"antigravity-claude-opus-4-6-thinking": {
"name": "Claude Opus 4.6 Thinking (Antigravity)",
"limit": {
"context": 200000,
"output": 64000
},
"modalities": {
"input": [
"text",
"image",
"pdf"
],
"output": [
"text"
]
}
},
"antigravity-gpt-oss-120b-medium": {
"name": "GPT-OSS 120B Medium (Antigravity)",
"limit": {
"context": 200000,
"output": 64000
},
"modalities": {
"input": [
"text",
"image",
"pdf"
],
"output": [
"text"
]
}
},
"antigravity-gemini-3.5-flash-low": {
"name": "Gemini 3.5 Flash Low (Antigravity)",
"limit": {
"context": 1048576,
"output": 65536
},
"modalities": {
"input": [
"text",
"image",
"pdf"
],
"output": [
"text"
]
}
},
"antigravity-gemini-3.5-flash-high": {
"name": "Gemini 3.5 Flash High (Antigravity)",
"limit": {
"context": 1048576,
"output": 65536
},
"modalities": {
"input": [
"text",
"image",
"pdf"
],
"output": [
"text"
]
}
},
"antigravity-gemini-3.5-pro-low": {
"name": "Gemini 3.5 Pro Low (Antigravity)",
"limit": {
"context": 1048576,
"output": 65536
},
"modalities": {
"input": [
"text",
"image",
"pdf"
],
"output": [
"text"
]
}
},
"antigravity-gemini-3.5-pro-high": {
"name": "Gemini 3.5 Pro High (Antigravity)",
"limit": {
"context": 1048576,
"output": 65536
},
"modalities": {
"input": [
"text",
"image",
"pdf"
],
"output": [
"text"
]
}
}
}
}
}
}Backward Compatibility: Legacy model names still resolve in the plugin where possible, but they are no longer listed by default. Prefer the current Antigravity quota model IDs above.
Multi-Account Setup
Add multiple Google accounts for a higher combined quota. The plugin automatically rotates between accounts when one is rate-limited.
opencode auth login # Run again to add more accountsAccount management options (via opencode auth login):
- Configure models — Auto-configure all plugin models in opencode.json
- Check quotas — View remaining API quota for each account
- Manage accounts — Enable/disable specific accounts for rotation
For details on load balancing, dual quota pools, and account storage, see docs/MULTI-ACCOUNT.md.
Troubleshooting
Quick Reset: Most issues can be resolved by deleting
~/.config/opencode/antigravity-accounts.jsonand runningopencode auth loginagain.
Configuration Path (All Platforms)
OpenCode uses ~/.config/opencode/ on all platforms including Windows.
| File | Path |
|------|------|
| Main config | ~/.config/opencode/opencode.json |
| Accounts | ~/.config/opencode/antigravity-accounts.json |
| Plugin config | ~/.config/opencode/antigravity.json |
| Debug logs | ~/.config/opencode/antigravity-logs/ |
Windows users:
~resolves to your user home directory (e.g.,C:\Users\YourName). Do NOT use%APPDATA%.
Custom path: Set
OPENCODE_CONFIG_DIRenvironment variable to use a custom location.
Windows migration: If upgrading from plugin v1.3.x or earlier, the plugin will automatically find your existing config in
%APPDATA%\opencode\and use it. New installations use~/.config/opencode/.
Multi-Account Auth Issues
If you encounter authentication issues with multiple accounts:
- Delete the accounts file:
rm ~/.config/opencode/antigravity-accounts.json - Re-authenticate:
opencode auth login
403 Permission Denied (rising-fact-p41fc)
Error:
Permission 'cloudaicompanion.companions.generateChat' denied on resource
'//cloudaicompanion.googleapis.com/projects/rising-fact-p41fc/locations/global'Cause: Plugin falls back to a default project ID when no valid project is found. This works for Antigravity but fails for Gemini CLI models.
Solution:
- Go to Google Cloud Console
- Create or select a project
- Enable the Gemini for Google Cloud API (
cloudaicompanion.googleapis.com) - Add
projectIdto your accounts file:{ "accounts": [ { "email": "[email protected]", "refreshToken": "...", "projectId": "your-project-id" } ] }
Note: Do this for each account in a multi-account setup.
Gemini Model Not Found
Add this to your google provider config:
{
"provider": {
"google": {
"npm": "@ai-sdk/google",
"models": { ... }
}
}
}Gemini 3 Models 400 Error ("Unknown name 'parameters'")
Error:
Invalid JSON payload received. Unknown name "parameters" at 'request.tools[0]'Causes:
- Tool schema incompatibility with Gemini's strict protobuf validation
- MCP servers with malformed schemas
- Plugin version regression
Solutions:
Update to latest beta:
{ "plugin": ["opencode-antigravity-auth@beta"] }Disable MCP servers one-by-one to find the problematic one
Add npm override:
{ "provider": { "google": { "npm": "@ai-sdk/google" } } }
MCP Servers Causing Errors
Some MCP servers have schemas incompatible with Antigravity's strict JSON format.
Common symptom:
Invalid function name must start with a letter or underscoreSometimes it shows up as:
GenerateContentRequest.tools[0].function_declarations[12].name: Invalid function name must start with a letter or underscoreThis usually means an MCP tool name starts with a number (for example, a 1mcp key like 1mcp_*). Rename the MCP key to start with a letter (e.g., gw) or disable that MCP entry for Antigravity models.
Diagnosis:
- Disable all MCP servers in your config
- Enable one-by-one until error reappears
- Report the specific MCP in a GitHub issue
"All Accounts Rate-Limited" (But Quota Available)
Cause: Cascade bug in clearExpiredRateLimits() in hybrid mode (fixed in recent beta).
Solutions:
- Update to latest beta version
- If persists, delete accounts file and re-authenticate
- Try switching
account_selection_strategyto"sticky"inantigravity.json
Session Recovery
If you encounter errors during a session:
- Type
continueto trigger the recovery mechanism - If blocked, use
/undoto revert to pre-error state - Retry the operation
Using with Oh-My-OpenCode
Important: Disable the built-in Google auth to prevent conflicts:
// ~/.config/opencode/oh-my-opencode.json
{
"google_auth": false,
"agents": {
"frontend-ui-ux-engineer": { "model": "google/antigravity-gemini-3.1-pro-high" },
"document-writer": { "model": "google/antigravity-gemini-3.5-flash-low" }
}
}Infinite .tmp Files Created
Cause: When account is rate-limited and plugin retries infinitely, it creates many temp files.
Workaround:
- Stop OpenCode
- Clean up:
rm ~/.config/opencode/*.tmp - Add more accounts or wait for rate limit to expire
OAuth Callback Issues
Symptoms:
- "fail to authorize" after successful Google login
- Safari shows "Safari can't open the page"
Cause: Safari's "HTTPS-Only Mode" blocks http://localhost callback.
Solutions:
Use Chrome or Firefox (easiest): Copy the OAuth URL and paste into a different browser.
Disable HTTPS-Only Mode temporarily:
- Safari > Settings (⌘,) > Privacy
- Uncheck "Enable HTTPS-Only Mode"
- Run
opencode auth login - Re-enable after authentication
macOS / Linux:
# Find process using the port
lsof -i :51121
# Kill if stale
kill -9 <PID>
# Retry
opencode auth loginWindows (PowerShell):
netstat -ano | findstr :51121
taskkill /PID <PID> /F
opencode auth loginOAuth callback requires browser to reach localhost on the machine running OpenCode.
WSL2:
- Use VS Code's port forwarding, or
- Configure Windows → WSL port forwarding
SSH / Remote:
ssh -L 51121:localhost:51121 user@remoteDocker / Containers:
- OAuth with localhost redirect doesn't work in containers
- Wait 30s for manual URL flow, or use SSH port forwarding
Configuration Key Typo: plugin not plugins
The correct key is plugin (singular):
{
"plugin": ["opencode-antigravity-auth@beta"]
}Not "plugins" (will cause "Unrecognized key" error).
Migrating Accounts Between Machines
When copying antigravity-accounts.json to a new machine:
- Ensure the plugin is installed:
"plugin": ["opencode-antigravity-auth@beta"] - Copy
~/.config/opencode/antigravity-accounts.json - If you get "API key missing" error, the refresh token may be invalid — re-authenticate
Known Plugin Interactions
For details on load balancing, dual quota pools, and account storage, see docs/MULTI-ACCOUNT.md.
Plugin Compatibility
@tarquinen/opencode-dcp
DCP creates synthetic assistant messages that lack thinking blocks. List this plugin BEFORE DCP:
{
"plugin": [
"opencode-antigravity-auth@latest",
"@tarquinen/opencode-dcp@latest"
]
}oh-my-opencode
Disable built-in auth and override agent models in oh-my-opencode.json:
{
"google_auth": false,
"agents": {
"frontend-ui-ux-engineer": { "model": "google/antigravity-gemini-3.1-pro-high" },
"document-writer": { "model": "google/antigravity-gemini-3.5-flash-low" },
"multimodal-looker": { "model": "google/antigravity-gemini-3.5-flash-low" }
}
}Tip: When spawning parallel subagents, enable
pid_offset_enabled: trueinantigravity.jsonto distribute sessions across accounts.
Plugins you don't need
- gemini-auth plugins — Not needed. This plugin handles all Google OAuth.
Configuration
Create ~/.config/opencode/antigravity.json for optional settings:
{
"$schema": "https://raw.githubusercontent.com/NoeFabris/opencode-antigravity-auth/main/assets/antigravity.schema.json"
}Most users don't need to configure anything — defaults work well.
Model Behavior
| Option | Default | What it does |
|--------|---------|--------------
| keep_thinking | false | Preserve Claude's thinking across turns. Warning: enabling may degrade model stability. |
| session_recovery | true | Auto-recover from tool errors |
| cli_first | false | Legacy option for Gemini CLI quota fallback. Current default model definitions avoid Gemini CLI because individual access sunsets on 2026-06-18. |
Account Rotation
| Your Setup | Recommended Config |
|------------|-------------------|
| 1 account | "account_selection_strategy": "sticky" |
| 2-5 accounts | Default ("hybrid") works great |
| 5+ accounts | "account_selection_strategy": "round-robin" |
| Parallel agents | Add "pid_offset_enabled": true |
Quota Protection
| Option | Default | What it does |
|--------|---------|--------------|
| soft_quota_threshold_percent | 90 | Skip account when quota usage exceeds this percentage. Prevents Google from penalizing accounts that fully exhaust quota. Set to 100 to disable. |
| quota_refresh_interval_minutes | 15 | Background quota refresh interval. After successful API requests, refreshes quota cache if older than this interval. Set to 0 to disable. |
| soft_quota_cache_ttl_minutes | "auto" | How long quota cache is considered fresh. "auto" = max(2 × refresh interval, 10 minutes). Set a number (1-120) for fixed TTL. |
How it works: Quota cache is refreshed automatically after API requests (when older than
quota_refresh_interval_minutes) and manually via "Check quotas" inopencode auth login. The threshold check usessoft_quota_cache_ttl_minutesto determine cache freshness - if cache is older, the account is considered "unknown" and allowed (fail-open). When ALL accounts exceed the threshold, the plugin waits for the earliest quota reset time (like rate limit behavior). If wait time exceedsmax_rate_limit_wait_seconds, it errors immediately.
Rate Limit Scheduling
Control how the plugin handles rate limits:
| Option | Default | What it does |
|--------|---------|--------------|
| scheduling_mode | "cache_first" | "cache_first" = wait for same account (preserves prompt cache), "balance" = switch immediately, "performance_first" = round-robin |
| max_cache_first_wait_seconds | 60 | Max seconds to wait in cache_first mode before switching accounts |
| failure_ttl_seconds | 3600 | Reset failure count after this many seconds (prevents old failures from permanently penalizing accounts) |
When to use each mode:
- cache_first (default): Best for long conversations. Waits for the same account to recover, preserving your prompt cache.
- balance: Best for quick tasks. Switches accounts immediately when rate-limited for maximum availability.
- performance_first: Best for many short requests. Distributes load evenly across all accounts.
App Behavior
| Option | Default | What it does |
|--------|---------|--------------|
| quiet_mode | false | Hide toast notifications |
| debug | false | Enable debug file logging (~/.config/opencode/antigravity-logs/) |
| debug_tui | false | Show debug logs in the TUI log panel (independent from debug) |
| auto_update | true | Auto-update plugin |
Transport (Advanced)
The plugin supports three transport modes. The default gateway mode is recommended for all normal use.
| Mode | Default | Description |
|------|---------|-------------|
| gateway | ✅ | CloudCode gateway shim — full streaming, tool calls, multi-account |
| cli | No | Opt-in: runs agy --print as subprocess. Text only, no tool calls. |
| managed-agent | No | Opt-in: public Managed Agents API with Gemini API key. Billing applies. |
⚠️ Gemini CLI sunset: 2026-06-18. The
gemini-cliquota pool stops serving requests for Google AI Pro/Ultra and free Gemini Code Assist individuals after this date. If requests fail after June 18, disablequota_fallbackin your config.
See docs/SUPPORT_MATRIX.md for the full feature/model/credential matrix per transport.
For all options, see docs/CONFIGURATION.md.
Environment variables:
OPENCODE_CONFIG_DIR=/path/to/config opencode # Custom config directory
OPENCODE_ANTIGRAVITY_DEBUG=1 opencode # Enable debug file logging
OPENCODE_ANTIGRAVITY_DEBUG=2 opencode # Verbose debug file logging
OPENCODE_ANTIGRAVITY_DEBUG_TUI=1 opencode # Enable TUI log panel debug outputTroubleshooting
See the full Troubleshooting Guide for solutions to common issues including:
- Auth problems and token refresh
- "Model not found" errors
- Session recovery
- Gemini CLI permission errors
- Safari OAuth issues
- Plugin compatibility
- Migration guides
Documentation
- Configuration — All configuration options
- Multi-Account — Load balancing, dual quota pools, account storage
- Model Variants — Thinking budgets and variant system
- Troubleshooting — Common issues and fixes
- Architecture — How the plugin works
- API Spec — Antigravity API reference
- Support Matrix — Feature/model/credential matrix per transport
Support
If this plugin saves you time, consider supporting its development:
Credits
License
MIT License. See LICENSE for details.
Intended Use
- Personal / internal development only
- Respect internal quotas and data handling policies
- Not for production services or bypassing intended limits
Warning
By using this plugin, you acknowledge:
- Terms of Service risk — This approach may violate ToS of AI model providers
- Account risk — Providers may suspend or ban accounts
- No guarantees — APIs may change without notice
- Assumption of risk — You assume all legal, financial, and technical risks
Disclaimer
- Not affiliated with Google. This is an independent open-source project.
- "Antigravity", "Gemini", "Google Cloud", and "Google" are trademarks of Google LLC.
