Antigravity Auth Configuration Options Complete Reference Manual

What You'll Learn

Find and modify all configuration options for the Antigravity Auth plugin
Understand the purpose and use cases for each configuration item
Choose the best configuration combination based on your usage scenario
Override configuration file settings using environment variables

Core Approach

The Antigravity Auth plugin controls almost all behavior through configuration files: from log level to account selection strategy, from session recovery to token refresh mechanism.

Configuration File Locations (priority from high to low)

Project Configuration: .opencode/antigravity.json
User Configuration:
- Linux/Mac: ~/.config/opencode/antigravity.json
- Windows: %APPDATA%\opencode\antigravity.json

Environment Variable Priority

All configuration items can be overridden via environment variables. Environment variables have higher priority than configuration files.

Configuration Overview

Category	Configuration Items	Core Scenarios
General Settings	3	Logging, debug mode
Thinking Block	1	Preserve thinking process
Session Recovery	3	Automatic error recovery
Signature Cache	4	Thinking block signature persistence
Empty Response Retry	2	Handle empty responses
Tool ID Recovery	1	Tool matching
Tool Hallucination Prevention	1	Prevent parameter errors
Token Refresh	3	Proactive refresh mechanism
Rate Limiting	5	Account rotation and waiting
Health Score	7	Hybrid strategy scoring
Token Bucket	3	Hybrid strategy tokens
Auto Update	1	Plugin auto-update
Web Search	2	Gemini search

General Settings

`quiet_mode`

Type: booleanDefault: falseEnvironment Variable: OPENCODE_ANTIGRAVITY_QUIET=1

Suppress most toast notifications (rate limits, account switching, etc.). Recovery notifications (successful session recovery) are always displayed.

Use Cases:

Multi-account high-frequency usage scenarios to avoid frequent notification interference
Automated scripts or background services

Example:

json

{
  "quiet_mode": true
}

`debug`

Type: booleanDefault: falseEnvironment Variable: OPENCODE_ANTIGRAVITY_DEBUG=1

Enable debug logging to file. Log files are stored by default in ~/.config/opencode/antigravity-logs/.

Use Cases:

Enable when troubleshooting issues
Provide detailed logs when submitting bug reports

Debug logs may contain sensitive information

Log files contain API responses, account indexes, and other information. Please sanitize before submitting.

`log_dir`

Type: stringDefault: OS-specific config directory + /antigravity-logsEnvironment Variable: OPENCODE_ANTIGRAVITY_LOG_DIR=/path/to/logs

Custom directory for storing debug logs.

Use Cases:

Need to store logs in a specific location (e.g., network shared directory)
Log rotation and archiving scripts

Thinking Block Settings

`keep_thinking`

Type: booleanDefault: falseEnvironment Variable: OPENCODE_ANTIGRAVITY_KEEP_THINKING=1

Experimental Feature

Preserve Claude model thinking blocks (via signature cache).

Behavior:

false (default): Strip thinking blocks to avoid signature errors, prioritizing reliability
true: Preserve complete context (including thinking blocks), but may encounter signature errors

Use Cases:

Need to view the model's complete reasoning process
Frequently use thinking content in conversations

Not Recommended For:

Production environments (reliability priority)
Multi-turn conversations (easily trigger signature conflicts)

Use with signature_cache

When enabling keep_thinking, it's recommended to configure signature_cache to improve signature hit rate.

Session Recovery

`session_recovery`

Type: booleanDefault: true

Automatically recover sessions from tool_result_missing errors. When enabled, toast notifications are displayed when recoverable errors occur.

Recoverable Error Types:

tool_result_missing: Tool result missing (ESC interrupt, timeout, crash)
Expected thinking but found text: Thinking block order error

Use Cases:

All scenarios using tools (default recommendation: enable)
Long conversations or frequent tool execution

`auto_resume`

Type: booleanDefault: false

Automatically send "continue" prompt to recover session. Only takes effect when session_recovery is enabled.

Behavior:

false: Only show toast notification, user needs to manually send "continue"
true: Automatically send "continue" to continue session

Use Cases:

Automated scripts or unattended scenarios
Want fully automated recovery process

Not Recommended For:

Need manual confirmation of recovery results
Need to check state before continuing after tool interruption

`resume_text`

Type: stringDefault: "continue"

Custom text sent during automatic recovery. Only used when auto_resume is enabled.

Use Cases:

Multilingual environments (e.g., change to "继续", "请继续")
Scenarios requiring additional prompt words

Example:

json

{
  "auto_resume": true,
  "resume_text": "请继续完成之前的任务"
}

Signature Cache

Only takes effect when keep_thinking is enabled

`signature_cache.enabled`

Type: booleanDefault: true

Enable disk caching of thinking block signatures.

Purpose: Caching signatures can avoid errors caused by repeated signing in multi-turn conversations.

`signature_cache.memory_ttl_seconds`

Type: number (range: 60-86400) Default: 3600 (1 hour)

Expiration time for memory cache (seconds).

`signature_cache.disk_ttl_seconds`

Type: number (range: 3600-604800) Default: 172800 (48 hours)

Expiration time for disk cache (seconds).

`signature_cache.write_interval_seconds`

Type: number (range: 10-600) Default: 60

Interval for background writing to disk (seconds).

Example:

json

{
  "keep_thinking": true,
  "signature_cache": {
    "enabled": true,
    "memory_ttl_seconds": 7200,
    "disk_ttl_seconds": 259200,
    "write_interval_seconds": 120
  }
}

Empty Response Retry

Automatically retry when Antigravity returns an empty response (no candidates/choices).

`empty_response_max_attempts`

Type: number (range: 1-10) Default: 4

Maximum number of retry attempts.

`empty_response_retry_delay_ms`

Type: number (range: 500-10000) Default: 2000 (2 seconds)

Delay between each retry (milliseconds).

Use Cases:

Unstable network environments (increase retry count)
Need fast failure (reduce retry count and delay)

Tool ID Recovery

`tool_id_recovery`

Type: booleanDefault: true

Enable tool ID orphan recovery. When tool response IDs don't match (due to context compression), try matching by function name or creating placeholders.

Purpose: Improve reliability of tool calls in multi-turn conversations.

Use Cases:

Long conversation scenarios (recommended to enable)
Scenarios with frequent tool usage

Tool Hallucination Prevention

`claude_tool_hardening`

Type: booleanDefault: true

Enable tool hallucination prevention for Claude models. When enabled, automatically inject:

Parameter signatures into tool descriptions
Strict tool usage rules system instruction

Purpose: Prevent Claude from using parameter names from training data instead of actual schema.

Use Cases:

Using MCP plugins or custom tools (recommended to enable)
Complex tool schemas

Not Recommended For:

Confirmed tool calls fully comply with schema (can disable to reduce extra prompts)

Proactive Token Refresh

`proactive_token_refresh`

Type: booleanDefault: true

Enable proactive background token refresh. When enabled, tokens are automatically refreshed before expiration, ensuring requests won't be blocked by token refresh.

Purpose: Avoid request delays waiting for token refresh.

`proactive_refresh_buffer_seconds`

Type: number (range: 60-7200) Default: 1800 (30 minutes)

How long before token expiration to trigger proactive refresh (seconds).

`proactive_refresh_check_interval_seconds`

Type: number (range: 30-1800) Default: 300 (5 minutes)

Interval for proactive refresh checks (seconds).

Use Cases:

High-frequency request scenarios (recommended to enable proactive refresh)
Want to reduce refresh failure risk (increase buffer time)

Rate Limiting and Account Selection

`max_rate_limit_wait_seconds`

Type: number (range: 0-3600) Default: 300 (5 minutes)

Maximum wait time (seconds) when all accounts are rate-limited. If the minimum wait time for all accounts exceeds this threshold, the plugin will fail fast instead of hanging.

Set to 0: Disable timeout, wait indefinitely.

Use Cases:

Scenarios requiring fast failure (reduce wait time)
Scenarios accepting long waits (increase wait time)

`quota_fallback`

Type: booleanDefault: false

Enable quota fallback for Gemini models. When the preferred quota pool (Gemini CLI or Antigravity) is exhausted, try the backup quota pool for the same account.

Use Cases:

High-frequency usage of Gemini models (recommended to enable)
Want to maximize quota utilization per account

Only takes effect when quota suffix is not explicitly specified

If the model name explicitly contains :antigravity or :gemini-cli, the specified quota pool will always be used without fallback.

`account_selection_strategy`

Type: string (enum: sticky, round-robin, hybrid) Default: "hybrid"Environment Variable: OPENCODE_ANTIGRAVITY_ACCOUNT_SELECTION_STRATEGY

Account selection strategy.

Strategy	Description	Use Cases
`sticky`	Use the same account until rate-limited, preserve prompt cache	Single session, cache-sensitive scenarios
`round-robin`	Rotate to the next account for each request, maximize throughput	Multi-account high-throughput scenarios
`hybrid`	Deterministic selection based on health score + token bucket + LRU	General recommendation, balances performance and reliability

For More Details

See Account Selection Strategies chapter.

`pid_offset_enabled`

Type: booleanDefault: falseEnvironment Variable: OPENCODE_ANTIGRAVITY_PID_OFFSET_ENABLED=1

Enable PID-based account offset. When enabled, different sessions (PIDs) will prioritize different starting accounts, helping distribute load when running multiple parallel agents.

Behavior:

false (default): All sessions start from the same account index, preserving Anthropic prompt cache (recommended for single session use)
true: Offset starting accounts based on PID, distribute load (recommended for multi-session parallel use)

Use Cases:

Running multiple parallel OpenCode sessions
Using sub-agents or parallel tasks

`switch_on_first_rate_limit`

Type: booleanDefault: true

Immediately switch accounts on first rate limit (after 1 second delay). When disabled, it will first retry the same account and only switch on the second rate limit.

Use Cases:

Want fast account switching (recommended to enable)
Want to maximize single account quota (disable)

Health Score (Hybrid Strategy)

Only takes effect when account_selection_strategy is hybrid

`health_score.initial`

Type: number (range: 0-100) Default: 70

Initial health score for accounts.

`health_score.success_reward`

Type: number (range: 0-10) Default: 1

Health score increase per successful request.

`health_score.rate_limit_penalty`

Type: number (range: -50-0) Default: -10

Health score deduction per rate limit.

`health_score.failure_penalty`

Type: number (range: -100-0) Default: -20

Health score deduction per failure.

`health_score.recovery_rate_per_hour`

Type: number (range: 0-20) Default: 2

Health score recovery per hour.

`health_score.min_usable`

Type: number (range: 0-100) Default: 50

Minimum health score threshold for account availability.

`health_score.max_score`

Type: number (range: 50-100) Default: 100

Health score upper limit.

Use Cases:

Default configuration works for most scenarios
High-frequency rate limit environments can reduce rate_limit_penalty or increase recovery_rate_per_hour
Need faster account switching can lower min_usable

Example:

json

{
  "account_selection_strategy": "hybrid",
  "health_score": {
    "initial": 80,
    "success_reward": 2,
    "rate_limit_penalty": -5,
    "failure_penalty": -15,
    "recovery_rate_per_hour": 5,
    "min_usable": 40,
    "max_score": 100
  }
}

Token Bucket (Hybrid Strategy)

Only takes effect when account_selection_strategy is hybrid

`token_bucket.max_tokens`

Type: number (range: 1-1000) Default: 50

Maximum capacity of the token bucket.

`token_bucket.regeneration_rate_per_minute`

Type: number (range: 0.1-60) Default: 6

Number of tokens generated per minute.

`token_bucket.initial_tokens`

Type: number (range: 1-1000) Default: 50

Initial number of tokens per account.

Use Cases:

High-frequency request scenarios can increase max_tokens and regeneration_rate_per_minute
Want faster account rotation can lower initial_tokens

Auto Update

`auto_update`

Type: booleanDefault: true

Enable plugin auto-update.

Use Cases:

Want to automatically get latest features (recommended to enable)
Need fixed version (disable)

Web Search (Gemini Grounding)

`web_search.default_mode`

Type: string (enum: auto, off) Default: "off"

Default mode for web search (when not specified via variant).

Mode	Description
`auto`	Model decides when to search (dynamic retrieval)
`off`	Search disabled by default

`web_search.grounding_threshold`

Type: number (range: 0-1) Default: 0.3

Dynamic retrieval threshold (0.0 to 1.0). Higher values mean the model searches less frequently (requires higher confidence to trigger search). Only takes effect in auto mode.

Use Cases:

Reduce unnecessary searches (increase threshold, e.g., 0.5)
Encourage model to search more (lower threshold, e.g., 0.2)

Example:

json

{
  "web_search": {
    "default_mode": "auto",
    "grounding_threshold": 0.4
  }
}

Configuration Examples

Single Account Basic Configuration

json

{
  "quiet_mode": false,
  "debug": false,
  "keep_thinking": false,
  "session_recovery": true,
  "auto_resume": false,
  "account_selection_strategy": "sticky"
}

Multi-Account High-Performance Configuration

json

{
  "quiet_mode": true,
  "debug": false,
  "session_recovery": true,
  "auto_resume": true,
  "account_selection_strategy": "hybrid",
  "quota_fallback": true,
  "switch_on_first_rate_limit": true,
  "max_rate_limit_wait_seconds": 120,
  "health_score": {
    "initial": 70,
    "min_usable": 40
  },
  "token_bucket": {
    "max_tokens": 100,
    "regeneration_rate_per_minute": 10
  }
}

Debugging and Diagnostics Configuration

json

{
  "debug": true,
  "log_dir": "/tmp/antigravity-logs",
  "quiet_mode": false,
  "session_recovery": true,
  "auto_resume": true,
  "tool_id_recovery": true
}

Preserve Thinking Block Configuration

json

{
  "keep_thinking": true,
  "signature_cache": {
    "enabled": true,
    "memory_ttl_seconds": 7200,
    "disk_ttl_seconds": 259200,
    "write_interval_seconds": 120
  },
  "session_recovery": true
}

FAQ

Q: How to temporarily disable a specific configuration?

A: Use environment variable override, no need to modify configuration file.

bash

# Temporarily enable debug mode
OPENCODE_ANTIGRAVITY_DEBUG=1 opencode

# Temporarily enable quiet mode
OPENCODE_ANTIGRAVITY_QUIET=1 opencode

Q: Do I need to restart OpenCode after modifying configuration file?

A: Yes, configuration file is loaded when OpenCode starts. You need to restart after modification.

Q: How to verify configuration takes effect?

A: Enable debug mode and check configuration loading information in log files.

json

{
  "debug": true
}

Logs will display loaded configuration:

[config] Loaded configuration: {...}

Q: Which configuration items most frequently need adjustment?

account_selection_strategy: Choose appropriate strategy for multi-account scenarios
quiet_mode: Reduce notification interference
session_recovery / auto_resume: Control session recovery behavior
debug: Enable when troubleshooting issues

Q: Is there JSON Schema validation for configuration file?

A: Yes, adding $schema field in configuration file enables IDE autocomplete and validation:

json

{
  "$schema": "https://raw.githubusercontent.com/NoeFabris/opencode-antigravity-auth/main/assets/antigravity.schema.json",
  ...
}

Appendix: Source Code Reference

Click to expand source code locations

Last updated: 2026-01-23

Feature	File Path	Lines
Configuration Schema Definition	`src/plugin/config/schema.ts`	1-373
JSON Schema	`assets/antigravity.schema.json`	1-157
Configuration Loading	`src/plugin/config/loader.ts`	-

Key Constants:

DEFAULT_CONFIG: Default configuration object (schema.ts:328-372)

Key Types:

AntigravityConfig: Main configuration type (schema.ts:322)
SignatureCacheConfig: Signature cache configuration type (schema.ts:323)
AccountSelectionStrategy: Account selection strategy type (schema.ts:22)

Antigravity Auth Configuration Options Complete Reference Manual ​

What You'll Learn ​

Core Approach ​

Configuration Overview ​

General Settings ​

quiet_mode ​

debug ​

log_dir ​

Thinking Block Settings ​

keep_thinking ​

Session Recovery ​

session_recovery ​

auto_resume ​

resume_text ​

Signature Cache ​

signature_cache.enabled ​

signature_cache.memory_ttl_seconds ​

signature_cache.disk_ttl_seconds ​

signature_cache.write_interval_seconds ​

Empty Response Retry ​

empty_response_max_attempts ​

empty_response_retry_delay_ms ​

Tool ID Recovery ​

tool_id_recovery ​

Tool Hallucination Prevention ​

claude_tool_hardening ​

Proactive Token Refresh ​

proactive_token_refresh ​

proactive_refresh_buffer_seconds ​

proactive_refresh_check_interval_seconds ​

Rate Limiting and Account Selection ​

max_rate_limit_wait_seconds ​

quota_fallback ​

account_selection_strategy ​

pid_offset_enabled ​

switch_on_first_rate_limit ​

Health Score (Hybrid Strategy) ​

health_score.initial ​

health_score.success_reward ​

health_score.rate_limit_penalty ​

health_score.failure_penalty ​

health_score.recovery_rate_per_hour ​

health_score.min_usable ​

health_score.max_score ​

Token Bucket (Hybrid Strategy) ​

token_bucket.max_tokens ​

token_bucket.regeneration_rate_per_minute ​

token_bucket.initial_tokens ​

Auto Update ​

auto_update ​

Web Search (Gemini Grounding) ​

web_search.default_mode ​

web_search.grounding_threshold ​

Configuration Examples ​

Single Account Basic Configuration ​

Multi-Account High-Performance Configuration ​

Debugging and Diagnostics Configuration ​

Preserve Thinking Block Configuration ​

FAQ ​

Q: How to temporarily disable a specific configuration? ​

Q: Do I need to restart OpenCode after modifying configuration file? ​

Q: How to verify configuration takes effect? ​

Q: Which configuration items most frequently need adjustment? ​

Q: Is there JSON Schema validation for configuration file? ​

Appendix: Source Code Reference ​

Antigravity Auth Configuration Options Complete Reference Manual

What You'll Learn

Core Approach

Configuration Overview

General Settings

`quiet_mode`

`debug`

`log_dir`

Thinking Block Settings

`keep_thinking`

Session Recovery

`session_recovery`

`auto_resume`

`resume_text`

Signature Cache

`signature_cache.enabled`

`signature_cache.memory_ttl_seconds`

`signature_cache.disk_ttl_seconds`

`signature_cache.write_interval_seconds`

Empty Response Retry

`empty_response_max_attempts`

`empty_response_retry_delay_ms`

Tool ID Recovery

`tool_id_recovery`

Tool Hallucination Prevention

`claude_tool_hardening`

Proactive Token Refresh

`proactive_token_refresh`

`proactive_refresh_buffer_seconds`

`proactive_refresh_check_interval_seconds`

Rate Limiting and Account Selection

`max_rate_limit_wait_seconds`

`quota_fallback`

`account_selection_strategy`

`pid_offset_enabled`

`switch_on_first_rate_limit`

Health Score (Hybrid Strategy)

`health_score.initial`

`health_score.success_reward`

`health_score.rate_limit_penalty`

`health_score.failure_penalty`

`health_score.recovery_rate_per_hour`

`health_score.min_usable`

`health_score.max_score`

Token Bucket (Hybrid Strategy)

`token_bucket.max_tokens`

`token_bucket.regeneration_rate_per_minute`

`token_bucket.initial_tokens`

Auto Update

`auto_update`

Web Search (Gemini Grounding)

`web_search.default_mode`

`web_search.grounding_threshold`

Configuration Examples

Single Account Basic Configuration

Multi-Account High-Performance Configuration

Debugging and Diagnostics Configuration

Preserve Thinking Block Configuration

FAQ

Q: How to temporarily disable a specific configuration?

Q: Do I need to restart OpenCode after modifying configuration file?

Q: How to verify configuration takes effect?

Q: Which configuration items most frequently need adjustment?

Q: Is there JSON Schema validation for configuration file?

Appendix: Source Code Reference