v0.2.2

Added

• 🤖 Per-model configuration. Set custom parameters (temperature, top_p, etc.) for each model or as global defaults. Per-chat overrides still take priority.
• 🧠 Automatic context compaction. Long conversations are automatically summarized to keep things running smoothly, no manual intervention needed.
• ➕ Per-chat parameters. Override model parameters for a single chat session from the + menu.

Changed

• ⚙️ Unified settings. Admin panel merged into Settings. No more separate modal. Admin sections appear automatically for admin users.
• ⌨️ Shortcut hint in menu. The Settings menu item now shows its keyboard shortcut.

v0.2.1

Added

• 🔍 Global search modal. New unified search (⌘K / ⌘⇧F) that finds chats by title and message content, plus files by name — across all workspaces or scoped to the active one. Shows recent chats when the query is empty. Replaces the old QuickOpen modal.
• 🌐 Perplexity web search provider. Added Perplexity as a first-class search provider (auto-detected between Exa and Tavily when PERPLEXITY_API_KEY is set).
• 🌐 Chat Completions search provider. Any OpenAI-compatible /chat/completions endpoint (e.g. Perplexity Sonar, LiteLLM proxy) can now be used for web search. Configure via Settings or environment variables.
• ➕ New Chat button per workspace. Each workspace row in the sidebar now shows a pencil icon on hover to quickly create a new chat.

Fixed

• 🔍 Search no longer pinned to last workspace. When on a non-workspace page (e.g. Automations), search results are now global instead of silently scoped to the last-selected workspace.
• 🔄 Sidebar chat cache invalidated globally. Chat events from automations or other tabs now clear the cache for all workspaces, so re-expanding any workspace shows fresh data.
• 🔄 New chats from automations appear in landing page. The chat landing now debounce-reloads when it receives a socket event for an unknown chat, so automation-created chats appear without a manual refresh.
• 🏷️ Chat tab labels update from DB title. Opening a chat from search or sidebar now shows the real title instead of a generic "Chat" label.
• 🔁 Duplicate new-chat tabs prevented. Opening a new chat reuses an existing empty/pending chat tab instead of creating duplicates.

Changed

• 🔗 URL-driven navigation for chat and file intents. Clicking a chat or file in the sidebar/search now uses URL query params (chatId, file, dir) which are consumed and cleaned up after navigation, enabling deep-linking and back-button support.
• 🍞 Improved toast notifications. Toaster now uses rich colors and close buttons for better visibility.
• 🔒 Automations sidebar link hidden when chat is disabled. The Automations nav item is now conditionally shown only when the chat/LLM backend is available.
• ⬆️ Python 3.10 minimum. Bumped requires-python from >=3.9 to >=3.10.
• 📦 Added truststore dependency. Uses platform-native TLS certificate stores for better certificate handling.
• 🌍 i18n for search UI. Search-related strings added across all 10 supported locales.

v0.2.0

Added

• ⏱️ Automations. Schedule recurring tasks that run on a timer. Each automation creates a real chat with full tool access, so they can do anything you'd do manually. Manage them from the new /automations page or ask the AI to create one.
• 🔔 Task completion notifications. Get a browser notification (with optional sound) when a background task finishes. Toggle in Settings > General.
• 🪝 Webhook notifications. Send alerts to Slack, Discord, Teams, or any webhook URL when tasks complete. Configure in Settings > General.
• 💬 Sidebar chat history. Each workspace now shows its recent chats directly in the sidebar. Click to reopen, updates live as new chats come in.
• 🗃️ Config file sync. Settings now persist to config.toml on disk, so your configuration survives across reinstalls and can be hand-edited.

Changed

• 🎨 Improved artifact cards. File previews in chat now show a file icon and cleaner styling.

v0.1.9

Fixed

• 📦 PyPI install actually works now. Hatchling's exclude took precedence over artifacts, causing every published wheel to ship without the frontend build. The app returned {"detail":"Not Found"} on all non-API routes. Switched to force-include (matching open-webui's approach), which unconditionally bundles cptr/frontend/build into the wheel regardless of exclude rules.

v0.1.8

Fixed

• 📦 PyPI install actually works now. Hatchling's exclude = ["cptr/frontend"] was taking precedence over the artifacts directive, causing every published wheel to ship without the frontend build. The app returned {"detail":"Not Found"} on all non-API routes. Changed to granular excludes that skip only source/dev files while preserving cptr/frontend/build.

v0.1.7

Fixed

• 🔄 Messages no longer disappear after sending. Fixed a race condition where the frontend needed a separate GET request after sending a message, creating a window where socket events were dropped. The POST now returns the created messages directly, eliminating the round-trip and the race.

v0.1.6

Added

• 📋 Plan mode. Request an implementation plan before AI changes.

Fixed

• 🔒 Protected .env files. File tools and search now block .env and .env.* access.

Changed

• ✨ Cleaner artifact cards. Simplified assistant artifact card headers and preview text.

v0.1.5

Fixed

• 🗄️ Alembic migration no longer fails on startup. Renamed migration revision IDs from 3-character (001) to 4-character (0001) identifiers, fixing Alembic's ResolutionError for partial revision matches.
• 📦 PyPI wheel no longer bundles frontend source code. Fixed .gitignore paths (old computr/ → cptr/) and added an explicit exclude in pyproject.toml so only cptr/frontend/build ships in the wheel — eliminating ~14,000 unnecessary files including node_modules.

v0.1.4

Fixed

• 🔄 Chat streaming no longer drops final content. Fixed a race where the done socket event arrived before the DB commit, causing a transient blank message. The streamed content now stays visible until the reload confirms it.
• 🔄 Unflushed text no longer lost at end of chat responses. The final text buffer is now properly flushed and emitted before marking a message as done, across all exit paths (normal completion, cancellation, and errors).
• 🔄 Intermediate chat state persisted during tool loops. Content and output items are now saved to the database between tool call iterations, so progress survives crashes or disconnects.
• 🧹 In-memory task state cleaned up on completion. The _task_state dict entry is now removed when a chat task finishes (done, cancelled, errored, or max iterations), preventing unbounded memory growth.
• 📱 Sidebar stays closed on mobile. The sidebar default breakpoint was raised to 1024px and an auto-close listener now collapses it whenever the viewport shrinks below 768px.
• 🔄 Stale chat loads discarded. Rapid chat switches no longer apply data from a slow earlier load, fixing a race condition that could show the wrong conversation.

Changed

• ⚡ All blocking filesystem I/O offloaded to threads. File reads, writes, directory walks, search, archive creation, uploads, renames, and deletions in the workspace and tool routers now run via asyncio.to_thread(), preventing event-loop stalls under heavy file operations.
• ⚡ Port scanner made fully async. Platform-specific port scanning (Darwin lsof, Linux /proc, Windows netstat) and PID-to-process lookups now use asyncio.create_subprocess_exec or asyncio.to_thread instead of blocking subprocess.run.
• ⚡ Welcome endpoint system info collected off the event loop. The /welcome handler now gathers hostname, memory, disk, CPU, network, and process data in a background thread.

v0.1.3

Fixed

• 📱 Tool calls no longer overflow on mobile. Fixed tool call rendering that could exceed the parent container width on narrow screens.
• 📱 Settings tabs scroll horizontally on mobile. The settings tab bar now scrolls on narrow screens instead of overflowing.

Changed

• 🔤 Tool call labels show filenames instead of full paths. Tool call summaries like Read, Edit, and Write now display just the filename (e.g. Read stores.ts) instead of the full absolute path, keeping labels readable on all screen sizes.
• 📋 Redesigned changelog modal. The changelog is now presented as a clean, continuous vertical list with color-coded section badges.