- Add independent Fable model mapping
- Rename startup model to default model
- Default to the Opus mapping and preserve Subagent inheritance
- Replace 1M checkboxes with standard toggle switches
- Align form fields and simplify explanatory text
- Remove bulk role synchronization
- Update translations and regression tests
- Add per-role model rows with inline 1M context controls
- Add bulk action to apply Sonnet settings to all roles
- Persist startup model and context settings in official settings.json fields
- Preserve custom role mappings and Subagent configuration
- Add multilingual labels and UI regression coverage
- Validate behavior with real Claude Code 2.1.292
Codex, Claude, and Gemini keys reject the disabled field, so toggling them made the whole config fail to load. Keep that field for OpenAI-compatible providers, disable native keys with excluded-models, and repair existing configs before the core starts.
Add independent controls for the startup model and Subagent model. Preserve Claude Code model selection and inheritance when CPA management is disabled.
Fixes#348
- Preserve hashed management keys written by the v8 core
- Avoid rewriting hashed keys with the GUI plaintext value
- Prevent duplicate top-level management mappings
Fixes#351
- Save pagination, visible columns, and column widths to the app data directory
- Migrate existing localStorage preferences and restore them on startup
- Preserve user choices instead of applying new defaults during upgrades
- Add regression tests for preference recovery and update preservation
Drop the separate reasoning-effort column from request details and render the raw effort, or auto, directly beneath the model name. Include it in the model tooltip, reset saved layouts that still use the old default column, and keep the field in CSV export.
- Preserve provider credentials, aliases, headers, and network settings during migration
- Normalize Claude cloak metadata and repair previously saved invalid types
- Use streaming Gemini health probes and honor group and key proxy settings
- Retain Codex customizations for temporarily unavailable models
- Add regression tests for migration, metadata, proxy routing, and catalog persistence
- Brighten usage subpages with consistent surfaces and text contrast
- Show reasoning effort by default and migrate saved column settings
- Hide HTTP method prefixes and tighten request table spacing
- Support custom Codex system prompts and synchronize instruction templates
- Add a clear button to plugin search
- Update localization, mocks, and regression tests
- Align response model handling with CPA-Manager-Plus and flag routing mismatches
- Add per-request costs and CSV export; fix mixed-context cost aggregation
- Display API-reported access types and move providers next to timestamps
- Use white table backgrounds and simplify cost and status displays
- Expand client discovery paths, validate executables, and prioritize installed clients
- Add model, pricing, client discovery, and UI regression tests
- Use native v8 routes for credentials, models, OAuth, and API calls
- Remove implicit legacy endpoint and response transformations
- Serialize OAuth exclusion updates while preserving other providers
- Send YAML as raw text and preserve explicit null request bodies
- Update mocks and regression tests, including migrated Claude aliases
- Reorganize settings cards and improve responsive layouts, save states, and navigation
- Move model aliases to a dedicated tab and improve search and keyboard interactions
- Add pagination to the agent client list and update localized management text
- Use public model IDs consistently for Codex selection and catalogs
- Fix v8 migration and preserve keys, providers, and aliases in mixed configurations
- Improve the stop-core button styling and expand UI and configuration regression tests
- Add plugin store, version selection, configuration, and OAuth login
- Support Muse (Meta) OAuth device authorization
- Redesign home dashboard and add batch model health checks
- Support per-credential provider editing, removal, and reordering
- Improve usage charts and track actual response models
- Preserve installed plugins and runtime data during portable rebuilds
- Update translations, documentation, and related tests
Keep core config writes on the watched file so repeated saves take effect.
Recover credentials from the update backup when the new macOS app has an empty OAuth directory.
Complete the v0.3.11 release notes in four languages.
Prefer diskutil eject with retries and a compatibility fallback. Defer failed cleanup without blocking verified updates, preserve mounted directories, and add regression tests.
- Extend OAuth directory recovery to Windows and Linux
- Align credential path resolution and log directory selection with the core
- Update v8 TLS and logging fields without requiring a version marker
- Support flow-style YAML mappings and preserve null legacy TLS fields
- Add regression tests for path recovery and config updates
- Normalize auth paths before comparison and credential scanning
- Update effective v8 auth-dir fields without requiring a version marker
- Preserve active credentials and configuration when recovery fails
- Add regression tests for config reloads and v7-to-v8 upgrades
- Default absent optional config nodes to empty values without hiding other errors
- Normalize empty parent mappings to prevent alias save and rollback conflicts
- Add regression tests for missing config, first-time alias creation, and rollback
Resolve Antigravity IDE configuration directory from installed executable metadata and product.json before falling back to legacy paths.
Launch Antigravity CLI via System.Diagnostics.Process and wait for process exit when running under Windows PowerShell to avoid console I/O contention.
Add unit tests for executable product metadata resolution and PowerShell CLI process synchronization.
- Add separate configuration entries for IDE and CLI
- Integrate Gemini-compatible routing and scoped credentials
- Support detection, launch, backup, restore, and IDE restart
- Add localized guidance and configuration, UI, and native integration tests