What shipped, version by version

Added

• Claude: Set up Claude directly from the Commit tool window when credentials are missing.
• DeepSeek: Added a dedicated Thinking Mode experience with live reasoning progress and reliable token statistics. Reasoning content is never inserted into commits, stored in history, or sent in telemetry.
• Diagnostics: Added a Copy diagnostic info action to make provider errors easier to troubleshoot.
• Model Catalog: View model context windows and continue browsing bundled model data when the remote catalog is unavailable.

Improved

• Settings: Redesigned provider configuration with dedicated pages, a clearer overview, consistent status indicators, inline reset actions, and clearer data-handling guidance.
• Commit Tool Window: Simplified the layout, moved provider selection to the title bar, made temperature easier to edit, and clarified preparing, reasoning, success, and error states.
• Accessibility: Improved keyboard navigation, focus behavior, contrast, responsive text wrapping, and light/dark theme consistency.
• Providers and Models: Updated default model lists, improved compatibility across supported providers, and made the model catalog faster and more reliable.

Fixed

• Commit Selection: Commit message generation now respects the changes selected for the current commit when launched from the AICommit tool window.
• Streaming: Incomplete responses now report an error instead of appearing successful; cancellation and provider or network failures are handled more reliably.
• Commit UI: Fixed generation and refresh issues across different commit workflows.
• Provider Setup: Temperature edits are saved before generation, and missing credentials now open the correct editable provider settings.
• Model Catalog: Fixed lost selections and clarified the difference between an empty catalog and no search matches.

Added

• DeepSeek: first-class provider with inline first-run setup.
• OpenAI-compatible: configure multiple profiles for custom endpoints and switch between them directly from the tool window.
• First-run: one-click local Ollama quick setup and contextual deep links from the tool window into the right Settings tab/field.
• History: commit-message history is now persisted per project.
• Telemetry: opt-out toggle in Settings with a privacy disclosure in About.
• UI: brand icons for built-in providers (OpenAI, Gemini, Azure, Claude, Ollama, DeepSeek) in the tool window provider dropdown and Settings provider tabs.
• UI: added the guarded provider settings navigation foundation.

Improved

• Onboarding: inline readiness guidance in the tool window after first success.
• Settings: internal navigation between the Provider and About sections.
• DeepSeek: refined Japanese and Korean copy.
• UI: converged the phase 1 design tokens.

Fixed

• Settings: removed an outer horizontal scrollbar and constrained the Data Handling privacy notice width.
• Commit Action: requires included changes at click time to avoid empty-selection generations.
• OpenAI-compatible: preserve profile references when Settings loads before profiles are fully populated.
• Telemetry: removed the IntelliJ internal `PluginManagerCore.getPlugin(PluginId)` call from plugin version reporting, resolving the Marketplace Internal API warning.

Added

• DeepSeek: first-class provider with inline first-run setup.
• OpenAI-compatible: configure multiple profiles for custom endpoints and switch between them directly from the tool window.
• First-run: one-click local Ollama quick setup and contextual deep links from the tool window into the right Settings tab/field.
• History: commit-message history is now persisted per project.
• Telemetry: opt-out toggle in Settings with a privacy disclosure in About.
• UI: brand icons for built-in providers (OpenAI, Gemini, Azure, Claude, Ollama, DeepSeek) in the tool window provider dropdown and Settings provider tabs.

Improved

• Onboarding: inline readiness guidance in the tool window after first success.
• Settings: internal navigation between the Provider and About sections.
• DeepSeek: refined Japanese and Korean copy.

Fixed

• Settings: removed an outer horizontal scrollbar and constrained the Data Handling privacy notice width.
• Commit Action: requires included changes at click time to avoid empty-selection generations.
• OpenAI-compatible: preserve profile references when Settings loads before profiles are fully populated.

Fixed

• OpenAI: preserve responses newline deltas.

Added

• Ollama: support for thinking models (e.g. DeepSeek-R1, kimi-k2.5) with adaptive context handling.
• Live streaming progress in the result panel with automatic scroll-follow.
• Settings: redesigned provider tabs with clearer active provider selection.
• Added Gemini model list; updated default models across providers.

Improved

• Standardized internal helper naming by removing misleading `ForTest` suffixes and keeping test-only helpers explicit.
• Reduced provider service trace noise and kept key mismatch/switch traces in local debug flow.
• Centralized provider capability decisions (temperature/role/token counting mode) under a shared entrypoint.
• Routed diff chunking branch selection through capability token counting mode while preserving provider-specific behavior.
• Made token metadata fallback decisions explicitly track source (`DIRECT/BACKEND/AZURE_OPENAI_FALLBACK/NONE`) for clearer diagnostics and tests.

Fixed

• `getCurrentAI()` now deterministically resolves to the configured provider from Settings.
• Unversioned file diff extraction now correctly handles deleted files and directories before text/binary branches.
• `stopGenerate()` now uses snapshot-then-clear to avoid losing cancellation on concurrently added jobs.

Added

• Added catalog-based model management to make provider/model selection more consistent.

Improved

• OpenAI now defaults to the `responses` endpoint mode.
• Improved stability for streaming responses and model list refresh.

Fixed

• Kept prompt drafts when switching providers or models.
• Prevented duplicate auto dialogs during a single interaction.
• Improved OpenAI error messages for clearer troubleshooting.
• Opened Settings directly when the API token is missing.
• Improved localization for model-manager and no-model messages.

Added

• Added catalog-based model management to make provider/model selection more consistent.

Improved

• OpenAI now defaults to the `responses` endpoint mode.
• Improved stability for streaming responses and model list refresh.

Fixed

• Kept prompt drafts when switching providers or models.
• Prevented duplicate auto dialogs during a single interaction.
• Improved OpenAI error messages for clearer troubleshooting.
• Opened Settings directly when the API token is missing.
• Improved localization for model-manager and no-model messages.

Added

• Added catalog-based model management to make provider/model selection more consistent.

Improved

• OpenAI now defaults to the `responses` endpoint mode.
• Improved stability for streaming responses and model list refresh.

Fixed

• Kept prompt drafts when switching providers or models.
• Prevented duplicate auto dialogs during a single interaction.
• Improved OpenAI error messages for clearer troubleshooting.
• Opened Settings directly when the API token is missing.
• Improved localization for model-manager and no-model messages.