Skip to content

CHANGELOG.md

Latest commit

 

History

History
275 lines (184 loc) · 10.7 KB

CHANGELOG.md

File metadata and controls

275 lines (184 loc) · 10.7 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.7.0] - 2026-03-02

Added

Enhanced Model Management

  • Dynamic Model Discovery: Per-key model fetching with 6-hour refresh cycle (MODEL_FETCH_INTERVAL=21600000)
    • Eliminates stale model errors (e.g., experimental models like gemini-2.0-flash-exp)
    • Request-driven discovery when model unknown
    • Automatic removal of models no longer available
  • Cross-Source Routing: Automatic failover across API keys → Antigravity OAuth → Gemini CLI OAuth
    • Configurable source preference order
    • Dynamic discovery across all sources
    • Unified model availability
  • Model Aliasing System: Create user-friendly aliases for frequently-used models
    • CLI commands: geminitro alias add/remove/list
    • Stored in .geminitro/models.json
    • Transparent resolution in request flow
    • Example: flash → gemini-2.0-flash
  • Quota Groups: Share quota limits across model variants
    • CLI commands: geminitro quota-group add/remove/list
    • Default groups for Gemini CLI and Antigravity models
    • When one model hits quota, entire group enters cooldown
  • Background Quota Refresh: Proactive 5-minute polling for OAuth keys (QUOTA_REFRESH_INTERVAL=300000)
    • Filters keys with <5% quota remaining before making requests
    • Prevents rate limit errors
    • OAuth-only feature (Antigravity, Gemini CLI)
  • Proper Case Normalization: Consistent "Gemini 2.0 Flash" formatting across all sources
    • User input normalized to Proper Case
    • Converted to kebab-case for API calls
    • Works with API keys, Antigravity OAuth, and Gemini CLI OAuth

New CLI Commands

  • geminitro alias add <name> <target> - Create model alias
  • geminitro alias remove <name> - Remove model alias
  • geminitro alias list - List all aliases
  • geminitro quota-group add <name> <models...> - Create quota group
  • geminitro quota-group remove <name> - Remove quota group
  • geminitro quota-group list - List all quota groups

New Configuration Options

  • MODEL_FETCH_INTERVAL - Model refresh interval (default: 6 hours)
  • QUOTA_REFRESH_INTERVAL - OAuth quota polling interval (default: 5 minutes)

New Services & Utilities

  • services/aliasService.js - Model aliasing CRUD operations
  • services/quotaGroupService.js - Quota group management
  • services/quotaRefreshService.js - Background quota polling service
  • src/cli/migrateModelsJson.js - Auto-migration script for models.json schema
  • utils/modelNormalizer.js - Proper Case ↔ kebab-case conversion utilities

Changed

  • Model Fetch Interval: Increased from 1 hour to 6 hours for better performance
  • models.json Schema: Migrated from array format to object with models, aliases, and quotaGroups
    • Auto-migration runs on server startup with timestamped backup
    • Fully backward compatible
  • Key Selection: Enhanced getOptimalKeyWithDiscovery() with cross-source routing logic
  • Model Normalization: All models displayed in Proper Case, sent to API in kebab-case

Fixed

  • Critical Bug: Removed thought_signature field from function calling (rejected by Gemini API in gemini-2.5-pro)
  • Stale Models: Dynamic discovery prevents "all keys exhausted" errors for experimental models
  • OAuth Fallback: Cross-source routing ensures OAuth sources are tried when API keys exhausted

Testing

  • Added 14 new unit tests (74 total across 13 test suites)
  • tests/keyService.crossSource.test.js - 7 tests for cross-source routing
  • tests/keyService.dynamicDiscovery.test.js - 7 tests for dynamic model discovery
  • All tests passing with 100% backward compatibility

Migration

Automatic - No manual action required. On first startup after upgrade:

  1. Server detects legacy models.json array format
  2. Creates timestamped backup (e.g., models.json.backup.1234567890)
  3. Migrates to new schema: { models: [...], aliases: {}, quotaGroups: {...} }
  4. Pre-configures default quota groups for Antigravity and Gemini CLI

Breaking Changes

None. Fully backward compatible with existing configurations.

1.5.5 - 2026-02-25

Changed

  • UI Improvements: Refined the log expansion toggle to appear as a "tab flap" anchored to the dark logs box for improved aesthetics and logical grouping.

1.5.4 - 2026-02-25

Changed

  • UI Improvements: Corrected the positioning of the "Expand/Collapse All" toggle in the dashboard logs to sit in the gap above the log box.

1.5.3 - 2026-02-25

Changed

  • UI Improvements: Repositioned the "Expand/Collapse All" (+) button in the dashboard logs to a floating position on the dark logs box for a cleaner layout.

Fixed

  • Settings Persistence: Implemented .env persistence for Verbose Logging and refactored settings updates for better reliability.

1.5.2 - 2026-02-25

Fixed

  • Account Persistence: Fixed issue where account types (Antigravity vs Gemini CLI) were not persisting after restart by adding source field to persistence layer.
  • Reliability: Optimized saveKeys to be synchronous to prevent race conditions during state updates.

1.5.1 - 2026-02-25

Fixed

  • Account Selection: Refactored getOptimalKey to strictly honor supportedModels and prevent fallback to accounts that explicitly exclude the requested model.

1.5.0 - 2026-02-25

Added

OAuth Authentication

  • Antigravity OAuth: Sign in with Google via OAuth for Antigravity accounts
  • Gemini CLI OAuth: Use existing Gemini CLI credentials with OAuth flow
  • Manual Token Entry: Support for manually entering refresh tokens via dashboard and CLI
  • Auto-import: Detect and import existing Antigravity accounts from OpenCode config
  • Browser-based Setup Wizard: Interactive first-run flow with OAuth options
  • CLI Commands: geminitro key oauth-antigravity, geminitro key oauth-gemini-cli, geminitro key import-antigravity

Model Discovery

  • Antigravity: 3-tier discovery strategy (loadCodeAssist API → GitHub spec → hardcoded fallback)
  • Gemini CLI: Direct retrieveUserQuota API for accurate model lists
  • Separate Logic: Each account type uses its own appropriate discovery method
  • Dynamic Refresh: Models are discovered and cached per account on addition

Enhanced Coding Agent Support

  • Tool Calls: Full function calling support with proper argument validation
  • Reasoning Effort: Maps OpenAI o-series reasoning_effort to Gemini thinkingConfig.budgetTokens
  • Extended Thinking: Supports Claude-style thinking.budget_tokens passthrough for Antigravity
  • Usage Telemetry: Accurate token usage tracking per request via stream_options.include_usage
  • Schema Validation: Strict JSON schema sanitization to prevent Gemini API "Unknown name" errors

Dashboard Improvements

  • Account Management: Visual distinction for API Key, Antigravity, and Gemini CLI accounts
  • Logo Assets: Display account type with appropriate logos (AI Studio, Antigravity, Gemini CLI)
  • OAuth Flow: In-dashboard OAuth initiation with state persistence and callback handling
  • Live Updates: Real-time account status updates via Socket.IO
  • Add Key Modal: Unified modal with OAuth options matching setup wizard design
  • Setup Wizard: OAuth options alongside traditional API key entry

New Services

  • services/antigravityService.js - Complete Antigravity/Gemini CLI API integration
  • services/oauthService.js - OAuth flow management, token exchange, refresh token handling

Changed

  • Key Pool: Now supports mixed account types (API keys + OAuth accounts)
  • Model Refresh: Separate discovery logic per account type instead of unified approach
  • Dashboard UI: "Add Key" button changed to "+ Add" for consistency
  • Setup Flow: Enhanced first-run experience with OAuth options

Fixed

  • Model discovery for Antigravity accounts now uses original 3-tier strategy (loadCodeAssist → GitHub → hardcoded)
  • Gemini CLI accounts no longer receive incorrect Antigravity model lists
  • JSON schema validation errors resolved by stripping unsupported keywords (const, default, pattern, $schema)
  • Tool call IDs now properly preserved and passed through to satisfy Antigravity/Claude backend requirements
  • Role-alternation grouping implemented to resolve payload validation issues
  • Strict allow-list for tool parameter schema keys to prevent "Unknown name" errors

Documentation

  • Added OAuth setup instructions to README
  • Updated .env.example with OAuth client credentials
  • Implementation plans in docs/plans/2026-02-23-oauth-authentication.md and docs/plans/2026-02-24-oauth-model-discovery.md

1.4.4 - 2026-02-24

Changed

  • Minor improvements and bug fixes

1.4.3 - 2026-02-23

Changed

  • Dashboard and agent setup enhancements

1.4.2 - 2026-02-22

Added

  • Agent registration improvements

1.4.1 - 2026-02-21

Added

  • Auto-detect coding agents with multi-select install

1.4.0 - 2026-02-20

Added

  • Major feature release with improved agent support

1.3.3 - 2026-02-19

Fixed

  • Multimodal content support improvements

1.3.2 - 2026-02-18

Fixed

  • Port conflict resolution

1.3.1 - 2026-02-17

Fixed

  • Auto-create .env and build dashboard on fresh install

1.3.0 - 2026-02-16

Added

  • Initial dashboard release

1.2.0 - 2026-02-15

Added

  • CLI improvements and key management