- Separate default model selection from catalog configuration
- Use API metadata with per-model and provider overrides
- Expose documented capability, reasoning, protocol, and request settings
- Preserve customizations with conflict detection and transactional rollback
- Discover installation paths from registry entries and shortcuts
- Handle installer records without InstallLocation
- Avoid launching the desktop app when version metadata is unavailable
- Add regression tests for path resolution and version detection
- Add configuration management tabs to full and easy modes
- Share run controls and feedback while preserving drafts and Claude settings
- Support desktop app restarts and Harness Web restarts with original parameters
- Improve responsive layouts, keyboard navigation, and localization
- Add UI regression and isolated process restart tests
Resolve model sources before replacing role aliases and support legacy
API and OAuth configurations that retain only aliased model entries.
Add regression coverage for source priority, disabled providers,
cross-role updates, and transaction rollback.
- Validate provider integrity before saving alias changes
- Reject rendered YAML that differs from the expected configuration
- Abort migration when existing config cannot be read or parsed
- Remove lossy config recovery fallback
- Add regression tests for provider loss, locked files, malformed config, and repeated updates
Do not treat managed aliases under disabled providers as usable Desktop routes. Match each route against its source on an enabled provider, skip disabled sources when creating aliases, and exclude disabled mappings when identifying direct aliases.
Validation: four regression tests fail before the fix and pass afterward; cargo test --locked (456 passed); cargo check --locked. After migrating the two affected live routes to the existing enabled source, Haiku and Opus messages requests both returned HTTP 200.
Use one cleanup path for override and override-raw across model and Fast alias creation and deletion. Remove supported effort and service-tier keys while preserving unrelated parameters, selector conditions, other protocols, and grouped models.
Validation: cargo test --locked alias_ (72 passed), including raw High/Fast deletion followed by Low/non-Fast recreation, orphaned raw rules, conditional grouped rules, and all supported effort keys.
Scope payload cleanup to the selected OAuth protocol and retain rules still used by surviving OAuth aliases, API mappings, or real models. Preserve protocol-free shared rules while any matching mapping survives, and limit creation cleanup to its source protocol.
Validation: cargo test --locked tests::alias_delete (6 passed), including Codex/Claude collisions, wildcard selectors, same-protocol OAuth channels, and API mappings.
Determine deletability from the actual provider model mapping instead of requiring an effort payload rule. Keep source models and unrelated aliases intact.
Validation: cargo test --locked tests::alias_delete (3 passed), covering all four API provider sections, removing every option in the editor, and real-model name collisions.
Include CPA model aliases in Desktop history previews and revision checks. Restore routes and local history through a serialized transaction, rolling core changes back when local restoration fails. Reject legacy fixed-route snapshots without mapping metadata.
Validation: cargo test --locked tests::alias_save (13 passed), including successful route restoration, stale previews, local conflicts, core write failure, and rollback.
Include Claude Desktop mappings in history change detection and preserve the previous selection. Show mapping changes in history previews and invalidate previews when metadata changes.
Validation: 15 history tests pass, including identical-profile model changes and repeated no-op saves.
Ignore inherited client configuration paths in test builds and enforce temporary-home boundaries for history fixtures.
Validation: 435 Rust tests pass, including a child-process regression with inherited client environment overrides.
- Validate configuration revisions and reject ambiguous model mappings
- Preserve raw overrides, conditions, and unchanged settings
- Serialize saves and restore configuration after write failures
- Add regression coverage and remove comments from affected files
Resolve existing aliases from their configuration, preserve metadata and unrelated payload parameters, and keep shared rules intact when renaming or changing sources.
Fix dialog feedback, keyboard focus, and narrow layouts. Add regression coverage for alias edits and browser interactions.
- Remove 3 unused functions and 2 unread fields
- Gate test-only types and helpers with #[cfg(test)]
- Verify all 403 Rust tests pass and check/release builds emit no warnings
- Retry errors 5, 32, 33, and 1175 up to five times with exponential backoff
- Preserve the final error and avoid retrying partial replacement failures
- Add eight regression tests covering transient and persistent file locks
Closes#241
- Stop live core processes before updates, regardless of port availability
- Wait for process termination before replacing files
- Block new core operations and cancel startup during shutdown
- Attach the Windows Job immediately after spawning
- Abort updates and restarts when stopping the core fails
- Add lifecycle and file-lock regression tests; all 33 related tests pass
Stop writing the OpenCode-specific npm field and clean up values
left by earlier versions when synchronizing ZCode configuration.
Add regression coverage for both app and CLI configs.
Closes#231
- Resolve context limits from core model definitions, aliases, and configuration
- Preserve user overrides and display context sources and fallback hints
- Refresh defaults from CPA when restoring all model settings
- Keep restore-all available without customizations and remove its icon
- Allow saving restored defaults even when values remain unchanged
- Add regression coverage for context resolution and default restoration
- Use consistent in-app dialogs for confirmation actions
- Keep quota refreshes free of prompts and automatic paid probes
- Add a separate confirmation flow for xAI availability tests
- Require explicit approval for Codex resets and prevent duplicate requests
- Show reset outcomes and require a quota refresh before retrying failures
- Cover confirmation and reset flows with fully mocked tests
- Use credential-file terminology in Simplified and Traditional Chinese
- Highlight the Enable button for disabled credentials without dimming actions
- Add regression tests for terminology and button styling
- Spawn Linux core processes on a dedicated, persistent thread instead of temporary background workers.
- Preserve parent-death signaling to stop the core when the GUI exits.
- Keep existing process management and Windows/macOS launch behavior unchanged.
- Add regression tests for launcher thread exit, runtime shutdown, spawn failure recovery, and parent-process termination.
- Move blocking usage queries and core lifecycle operations to background
workers, limit concurrency, and prevent overlapping core operations
- Coalesce statistics refreshes and consolidate category aggregations,
reducing unfiltered analysis queries from 11 to 4
- Cache bundled model prices, reuse management API connections, and reduce
redundant checks in the usage collector
- Throttle download progress and guide positioning updates, and avoid
redundant renders when core status is unchanged
- Bound and coalesce configuration watcher events, and reuse a single
model catalog refresh worker
- Stop scheduling remaining health checks when the dialog closes and
limit backend health-check concurrency
- Replace overlapping OAuth polling with sequential requests, retry
backoff, timeouts, and stale callback protection
- Add regression tests and a statistics benchmark script, and move
performance verification documentation to agent_docs
- add per-model context, reasoning, modality, visibility, and parallel tool settings
- persist customizations separately and apply them during catalog generation
- support restoring individual or all models to official fallback templates
- expose full fallback reasoning levels while keeping Fast available
- add localized UI, validation, documentation, and regression tests
- Sync all 11 dedicated model templates from the upstream Codex catalog
- Preserve official context windows, prompts, capabilities, and reasoning levels
- Remove the custom Spark and DeepSeek dedicated templates
- Move fallback model metadata into a standalone embedded resource
- Prevent remote catalogs and local caches from overriding fallback configuration
- Add catalog revision checks to prevent older templates from replacing newer ones
- Keep Fast mode available for both dedicated and fallback models
- Add regression coverage for Astra generation, catalog revisions, and fallback isolation
Validation: all 332 backend tests pass.
- Align fallback defaults and instruction structure with upstream Codex while preserving Fast availability for all models
- Stop forcing fallback reasoning levels or inferring original-image support
- Sync managed catalogs on model refresh, configuration changes, and every 30 seconds
- Remove unavailable models, replace missing defaults, and handle empty-list recovery
- Preserve existing configuration on request failures or invalid responses
- Serialize refreshes, update UI model lists, and preserve backups across restarts
- Add regression coverage for fallback compatibility, stale-model cleanup, and configuration recovery
- preserve the active Codex OAuth mode in legacy configuration paths
- retain unsaved OAuth selection when navigating between clients
- add regression coverage to prevent OAuth credentials from being overwritten
- add log diagnostics card under basic settings
- persist and synchronize CPA request-log configuration
- support configuration migration and rollback on save failure
- add localized UI copy and responsive styles
- add tests for request-log parsing and YAML merging
- use the original model slug as the catalog display name
- expose all standard reasoning levels for every model
- prevent upstream metadata from restricting reasoning options
- update catalog generation tests
- enable Fast for Codex API, OpenAI-compatible, and Codex OAuth sources
- apply service_tier=priority as an independent payload override
- support Fast-only and reasoning-effort combinations
- disable Fast until a supported model source is selected
lzt404