v0.4.8

Release v0.4.8

Fixed

• Preserve GPT-5.6 max reasoning effort in OpenAI provider options instead of dropping it during top-level reasoning translation.

v0.4.7

Release v0.4.7

Added

• Support GPT-5.6 max reasoning effort, reasoning mode and context.
• Add OpenAI Responses and Chat prompt-cache controls.
• Add Google standalone safety thresholds and Grok 4.5 dynamic discovery support.

Changed

• Update AI SDK core and provider packages to their latest patch releases.

v0.4.6

Release v0.4.6

Changed

• Harden context compaction with bounded rolling summaries, hard-ceiling enforcement, and lossless persisted overlays.
• Align AI SDK provider discovery, capabilities, options, reasoning, model selection, and caching.

Fixed

• Make compaction persistence and sub-agent control resilient to stale concurrent operations.
• Prevent stale chat history loads from overwriting active transcripts.

v0.4.5

Release v0.4.5

Added

• Providers: Grok (xAI) support, with API key and optional base URL credential fields and a new provider-request-headers layer for adapter-specific headers.

Changed

• Context: raised the compaction trigger threshold so summarization kicks in later, keeping more live context before compacting.

Fixed

• Chat: aligned composer popup glass surfaces (slash-command and mention menus) for a consistent look.

v0.4.4

Release v0.4.4

v0.4.3

v0.4.3

• Channels: multi-channel Chat SDK bridge connects the local agent to QQ,
  Discord, Lark, and WeChat — each an independent channel with its own adapter,
  encrypted (write-only) secret, and allowlist, sharing one agent and safety model.
• Browser: popup forwarding to new tabs, keyboard shortcuts, loading indicator,
  about:blank home with Bing search, and an address-bar editing guard.
• Providers: Zhipu and MiniMax support.
• Reasoning: unified into a single SDK-derived provider capability model routed
  through getReasoning IPC; composer badge reads provider defaults.
• MCP: reconnect-once and retry of remote requests on stale connections.
• CI/build: macOS hardened runtime, signing, and notarization via GitHub Secrets.
• Fixed: permission-badge flicker on conversation switch; Windows/POSIX path
  handling across plugin manifest, MCP cwd, and git/marketplace tests.

v0.4.2

v0.4.2

• Git sync returns a structured result so the UI reports real outcomes
  (received/published commits, conflicts, no-op) instead of silently refreshing.
• Git: discard partitions tracked vs. untracked paths (mixed batches no longer
  leave tracked files un-reverted); ahead/behind and upstream read via
  for-each-ref; git-sync-history-views split into git-sync-view and
  git-history-view; restore-confirm renamed to discard-confirm.
• Chat: long pasted text is externalized to a temp .txt file and handed to the
  model as a fileRead pointer, with the attachment chip kept in the transcript;
  new openPath/revealInFolder system IPC channels.
• App: refined update checks and reasoning defaults.

v0.4.1

v0.4.1

• Sub-agents: read-only drill-down into a sub-agent transcript (full-screen view
  with back button; header matches the app header and shows the agent name).
• Task rows: cancel, retry, and steer actions.
• Sub-agent scheduling hardened: dependents re-evaluated on every settle path,
  tasks kept alive on chat cancel, steer guarded against terminal and
  dependency-blocked tasks, approval waiters registered before the block is
  surfaced, structured dependency-failure tracking with unknown/failureKind
  feedback in the tools.
• Steer input schema no longer a top-level union.
• i18n: locale parity restored with missing zh-CN sub-agent keys.

v0.4.0

v0.4.0

• Context compaction v2: rebuilt subsystem with a ledger, injection, and a
  degrade chain; inline compaction integrated into the turn-loop, stream-runner,
  and coordinator; per-reply storage, message-steps types, and reworked context
  compile/sections/provider strategies. New per-step database migration.
• Goal v2: goal state machine with a cache-aware budget and a stable charter;
  aborted turns are now still accounted for.
• Provider: schema-driven call settings with per-conversation reasoning effort.
• Streaming performance: frame batching, split transcript stores, and hot-session
  reuse.
• Sidebar: pin conversations (pinned kept most-recently-pinned-first via a
  pinned_at timestamp; migration v26).
• Settings: add an About tab.
• Theme: extend code-font-size override to more components.

v0.3.0

v0.3.0

• Typography settings: interface-font + code-font pickers (bundled Geist, Inter,
  Geist Mono, JetBrains Mono; system fonts enumerated via the Local Font Access
  API), font-size slider (12–20 px), code-size slider (9–16 px), line-height
  slider (1.40–2.00), with a live preview panel. Replaces the old font-size preset.

• Window state persistence: the main window's position, size, and maximized state
  are saved across launches. Saved bounds are validated against the current
  displays to recover gracefully when a monitor is unplugged.

• Wallpaper system overhauled: a library of up to 12 stored assets with
  per-thumbnail previews, separate light/dark wallpaper selection (sun/moon
  badge), fit modes (cover / contain / fill / tile), and a surface/panel opacity
  slider controlling sidebar + toolbar + main-area transparency.

• Four @fontsource-variable packages added to bundle the typography fonts
  (Geist, Geist Mono, Inter, JetBrains Mono).

• Preferences schema: fontSizePresetId replaced by a typography object;
  WallpaperSettings expanded with assets, activeId, darkAssetId,
  fit, and surfaceOpacity.