February 2026

February 28, 2026

noktua

New Features

• Custom Agents in Settings — Added Settings > Agents panel with full CRUD capabilities for custom agents, including image, name, prompt, provider/model, thinking level, and tool selection
• Agent Selection in Chat — Model switcher now supports selecting custom agents from a dedicated Agents provider group with automatic workspace context capture on first use
• Runtime Agent Binding — Custom agents are bound at runtime with their configured prompt, workspace context, and tool allowlists, with agent prompts taking precedence over global personalization settings

Bug Fixes

• Thread Metadata Concurrency — Hardened thread metadata updates to prevent agentId from being dropped during concurrent operations
• Agent Message Rendering — Fixed values-only stream message parsing to ensure agent replies render correctly in the chat interface

New Features

• DiffMem scoped memory system — replaced Mem0 with DiffMem-based memory pipeline supporting project, agent, and global/user scopes across chat, settings, and tooling

Improvements

• Memory architecture — updated storage, IPC handlers, and UI components to integrate with new scoped memory system
• Documentation — refreshed memory-bank docs and contract checks for DiffMem migration

February 26, 2026

noktua

New Features

• macOS menu bar integration — Added native menu bar tray icon with click-to-toggle task popover for quick access to tasks and reminders

Improvements

• Tray icon rendering — Fixed invisible and square icon regressions by switching to macOS native named-template icon format
• Tray popover UX — Hidden traffic-light window controls for cleaner popover appearance
• Task management parity — Reused UserTasksContent component in tray popover for full CRUD and reminder functionality
• Tray IPC architecture — Implemented dedicated tray lifecycle management and IPC close channel with window.api.tray.close() preload API

February 25, 2026

noktua

New Features

• Connectors UX redesign — Settings → Connectors now uses a list-first master-detail flow matching the rest of Settings navigation patterns
• MCP stdio command validation — Introduced shared stdio command parser/validator utility with contract checks via check:mcp-stdio to prevent command-splitting and quoting regressions

Bug Fixes

• Quoted executable paths — Fixed handling of stdio commands with spaces to preserve re-parseable quoting and spawn reliably
• Task action parity — Corrected reminder controls to match hover/open visibility behavior consistently across Home and User Tasks surfaces

Improvements

• Connector visual hierarchy — Tuned UI with muted top actions (Test/Save), low-emphasis destructive action placement, and right-aligned outlined Add Connector CTA
• MCP error guidance — Enhanced ENOENT and stdio launch error messages to provide actionable troubleshooting instead of raw spawn errors

February 23, 2026

noktua

New Features

• IntegrationsPanel master-detail UI — Refactored to match ProvidersPanel pattern with typed detail components for each integration (BrowserUse, Morph, Figma, Perplexity, LangSmith, Mem0)
• Mem0 validation guards — API key now required before project ID input; toggle disabled without API key; added memoryModelId field to settings store and IPC
• Workspace memory modal — Unified modal for behavior prompt and mem0 project ID (replaces separate inline prompts)
• Workspace context menu — Added right-click context menu to workspace dropdown list items, matching active workspace header behavior

Improvements

• TypeScript integration contracts — New types/integrations.ts with typed detail props for all integrations
• Mem0 integration card visual parity — Now uses mem0.png logo and matches layout/styling of Perplexity/LangSmith cards
• Workspace memory menu UX — Consistent 'Memory' label, removed general-chat gating, improved right-click menu structure

Bug Fixes

• CI script compatibility — Replaced pnpm with npm in package.json scripts to fix CI runner failures
• Contract script macOS compatibility — Fixed grep -P to grep -E for macOS support
• WorkspaceAvatar context menu — Prevents native browser image context menu on right-click

New Features

• Workspace-scoped tasks — Tasks are now created and filtered per workspace; sidebar shows only active workspace tasks, HomeScreen displays latest 5 per selected workspace
• Inline task editing — Hover reveals edit, reminder, and delete actions; click pencil to edit in-place with Enter to save and Escape to cancel
• Task reminders with notifications — Set reminders via popover with 1-7 day selection and time picker; displays countdown (in Xd, in Xh, in Xm) that auto-refreshes every 60 seconds, with red "late" badge for overdue tasks
• macOS notification alerts — Reminder service fires Electron notifications with sound at scheduled times, respects global notification toggle, and persists across app restarts

Improvements

• Contextual workspace labels — Workspace tags show on hover in HomeScreen only; hidden in sidebar to reduce redundancy when already in workspace context
• Smart reminder cancellation — Reminders automatically cancel when tasks are completed or deleted

February 21, 2026

noktua

• Activity indicator on pinned workspaces — Pinned workspace tabs now display a pulsing blue dot when any thread is actively streaming or has running subagents
• gpt-image-1 compatibility fix — Fixed image generation by only setting response_format: "b64_json" for DALL-E models, as gpt-image-1 rejects this parameter
• Subagent manager stability — Fixed stale closure bug in updateSubagent callback and added 5-minute cleanup to free stuck concurrency slots
• Subagent lifecycle logging — Added debug logging to PerplexityRunner for better subagent tracking

New Features

• Automatic skill detection — Messages are now silently scored against installed skill triggers, auto-injecting the best-matching skill into the system prompt when no skill is manually selected
• LLM tiebreaker for ambiguous matches — Uses Anthropic Haiku with 3s timeout to resolve skill ambiguity when weighted scores fall within the confidence threshold zone
• Skill trigger configuration — Skills now support triggers[] and confidence_threshold fields in frontmatter for auto-detection tuning

Improvements

• Multi-signal skill scoring — Weighted keyword matching using exact (1.0), Porter stemming (0.7), substring (0.4), skill name (0.8), and description (0.2) scores
• System prompt safeguards — Added missing {{#section active_skill}} detection hint in Settings panel and updated prompt documentation

February 20, 2026

noktua

Bug Fixes

• Opus 4.6 fast mode silent failure — Detect "rate limit of 0" as fatal error to surface immediately instead of silently retrying for 3+ minutes

Improvements

• Gentler error banner styling — Replace harsh red error banner with softer warning-toned design using amber icon and neutral background

• Improved API plan error messaging — Add user-friendly error message for API plan/access issues when features like fast mode aren't available on current tier

• Dev mode runtime configuration — Added ~/.noktua/config.json toggle to enable developer settings, system prompt configuration, and PR tools, with separate dev-prompt.md for development use cases

Improvements

• Button and interactive element cursors — Added pointer cursor styling to all buttons, menus, options, and tabs for better UI consistency
• Simplified provider/integration status icons — Consolidated connected state to use a single CheckCircle2 icon to reduce visual clutter
• Commands panel refinements — Removed redundant / prefixes from inputs, added automatic kebab-casing, and made delete icons appear only on hover
• Chat empty state positioning — Moved initial "Ask Noktua" prompt lower for improved visual presentation
• Subagents sidebar icon update — Changed right sidebar tab icon from GitBranch to Bot for better clarity

Documentation

• Product roadmap added — Added three-phase roadmap covering Agent Rolodex & Enhanced Orchestration, Artifacts 2.0 (Universal Canvas), and Mac OS Integration (Nervous System)

February 19, 2026

noktua

New Features

• Notion-like document editor — Replaced Monaco code editor for .md files with a rich WYSIWYG editor powered by Novel.sh (Tiptap/ProseMirror)
• Slash command menu — Type / to insert headings, lists, code blocks, quotes, and dividers
• Floating bubble toolbar — Select text to access bold, italic, underline, strikethrough, code, and link formatting
• Link editing controls — Click link button to set/unset/edit URLs with inline input
• Auto-save with external change detection — Documents save automatically after 500ms and reload when AI edits the file
• AI chat context awareness — Open document file paths are automatically injected into AI message context

Bug Fixes

• Save status accuracy — Fixed bug where save status showed "saved" on write failure instead of "unsaved"
• Tiptap type compatibility — Resolved type mismatch between novel and tiptap-markdown dependency versions

Improvements

• Document tab behavior — Disabled rename for document tabs since titles are derived from file paths
• Dark mode support — Document editor now respects theme settings

New Features

• Hidden files in developer mode — File tree now displays dotfiles and hidden directories (e.g. .env, .git) when developer mode is enabled for a workspace

Bug Fixes

• Thread model preference — Threads now correctly use the global default model unless explicitly overridden per-thread
• General Chat workspace label — Removed chat bubble icon and changed "Personal" label to "General Chat" in workspace header and dropdown

New Features

• Notion-style document toolbar — Added persistent toolbar with headings, lists, quotes, code blocks, dividers, links, and undo/redo controls to the notes editor

Bug Fixes

• Autosave typing rollback fixed — Hardened document autosave and file-watch synchronization to prevent local typing from being overwritten during save operations
• Serialized note writes — Note saves are now serialized to prevent stale save completions from clobbering newer content

Improvements

• General Chat notes default to docs/ — New notes in the built-in General Chat workspace now default to docs/notes.md path, with ~/.noktua/chat/docs automatically created at startup while keeping other workspaces flexible

New Features

• Pinned workspace shortcut — Pin a workspace from the selector and access it quickly via an avatar in the title bar for instant switching

• Support up to 5 pinned workspaces — Pin multiple workspaces (up to 5) with avatars displayed in the titlebar; includes automatic migration from single-pin storage to array format with localStorage cleanup.

February 18, 2026

noktua

New Features

• Markdown file creation — Added ability to create new markdown files (.md) in the Files sidebar with filename validation

Improvements

• Refreshed Files sidebar UX — Replaced sync/download icon with clearer refresh icon and label, now triggers workspace reload

New Features

• Anthropic Claude 4.6 support — Added Claude Sonnet 4.6 with 1M context window and adaptive thinking mode
• Adaptive thinking mode — New reasoning option available for Anthropic 4.6 models, replacing budget mode from previous versions
• 1M context header — Automatically applies context-1m-2025-08-07 beta header for Sonnet 4.6 requests

Improvements

• Reasoning effort picker — Updated to show adaptive thinking option when using Anthropic 4.6 models

New Features

• Dev Mode worktrees — Added worktree management with IPC/service layer supporting list, create, remove, status, diff, start, stop, and open operations
• Developer settings panel — Added new settings panel with sidebar integration for worktree-aware thread workflows
• App lifecycle controls — Added start/stop controls in chat tabs with logs and progress surfaces in chat container

Bug Fixes

• ESLint compatibility — Fixed @eslint/js version alignment with ESLint 9 for CI compatibility

• Developer workspace settings UI redesign — Replaced single-workspace settings form with a workspace list and detail view, matching the Providers interaction pattern with per-workspace autosave (debounced, flushes on exit) and removed manual Save button

Bug Fixes

• Developer settings auto-save — Changed text fields to save on blur instead of real-time to prevent performance issues

Improvements

• Worktree creation flow — Replaced prompt-based UI with dialog-based flow including branch name normalization

• Worktree management actions — Added inline rename and delete actions with confirmation dialogs for improved UX

• Primary tab closing — Primary chat tabs can now be closed when multiple tabs exist; the next tab automatically becomes primary

• Anthropic Fast Mode toggle — Added a global Fast Mode setting in Settings → Providers → Anthropic that enables anthropic-beta: fast-mode-2026-02-01 header and speed: "fast" parameter for Claude Opus 4.6 requests

• Model switcher badge — Display a ⚡ lightning bolt indicator in the model switcher when Fast Mode is active and an Anthropic Opus 4.6 model is selected

• Personalization panel with custom instructions — Added a new settings panel that lets you write free-form instructions injected at the top of every agent system prompt; instructions are auto-saved locally to ~/.noktua/personalization.txt with 500ms debounce, and omitted from prompts when empty.

• General Chat workspace — Added a permanent built-in workspace pinned at the top of the sidebar, backed by ~/.noktua/chat/, providing a dedicated space for casual conversation not tied to any project folder with full agent file access. Protected against renaming, deletion, and pruning; auto-selected on first launch.

New Features

• Perplexity search tool — Added perplexity.search for quick web searches with citations via Perplexity sonar-pro (~5–10s response time)
• Perplexity deep research tool — Added perplexity.research for comprehensive multi-step research reports via sonar-deep-research (5–30 minute synthesis)
• Non-blocking research UI — Research tasks run in background sub-tabs with streaming results and automatic orchestrator notification on completion

Integrations

• Perplexity API key management — Added Perplexity integration card in Settings → Integrations for API key configuration

February 17, 2026

noktua

103 commits across 22 PRs — massive feature push across the entire app.

New Features

• Image generation sub-agent — dedicated tab view for AI image generation with a new sub-agent workflow
• Checkpoint timeline — fork and restore flow for conversation checkpoints, with hardened concurrent fork prevention
• Multi-model fanout — send a single prompt to multiple models simultaneously and compare responses side-by-side
• Plan mode — approval flow and execution handoff for agentic plan-then-execute workflows
• Canvas artifacts — rich artifact rendering with skills system and sub-agent support
• Mermaid diagrams — render mermaid diagrams inline in artifacts
• @ file tagging — tag files directly in the chat textarea for context-aware prompts
• Chat search — search previous conversations with jump-to-match navigation
• Slash commands — saved prompts accessible via / commands
• Queued messages — queue multiple messages with removable bubbles and batch sending modes
• macOS notifications — native notification settings and alerts
• Context usage meter — readable token usage bar with compaction indicators
• Long-chat resilience — automatic retries, context compaction, and graceful fallback for extended conversations
• File tree context menu — right-click actions on workspace files with auto-expand to depth 2
• Workspace-bound chat tabs — tabs are now scoped to the active workspace

Integrations & Providers

• OAuth for Claude Code & Codex — full OAuth flow replacing manual API key entry, with model catalog fetching
• Figma integration — Figma tool routing with /v1 API base path support
• Fireworks AI — added as a new provider option
• Cerebras — new provider with settings panel and model favorites
• LangSmith tracing — configuration UI in settings, moved to Integrations panel
• Browser Use — wired up settings and restored live sub-agent preview flow
• Tool registry — async background sub-agents with a centralized tool registry

Bug Fixes

• Fixed drag-and-drop file attachments stuck in loading state
• Fixed dropped image attachments without file paths
• Fixed agent tool input recovery for write_file/edit_file
• Fixed agent middleware ordering to prevent retry validation crash
• Corrected Codex model IDs (gpt-5.3-codex, gpt-5.3-codex-spark, gpt-5.2)
• Fixed OAuth redirect URI (localhost vs 127.0.0.1)
• Fixed Gemini token usage parsing for context bar
• Handled alternative token usage formats (ChatOpenAI, Cerebras)
• Fixed chat composer footer crowding and control layout
• Fixed duplicate tab bar and sidebar resize/collapse strip
• Fixed settings model list scrolling overflow
• Fixed theme persistence to localStorage (default to system)
• Fixed packaged Electron startup (missing ajv dep, deepagents bundling)
• Close open artifact when creating new thread
• Prevent selecting models from providers without API keys
• Apply enabled-models filter per-provider, not globally

UI Polish

• Chat bar polish — hidden EXECUTE badge, truncated model names, updated provider icons
• Sidebar redesign — flat agents panel, collapsible day groups, branding tweaks
• Providers panel — master-detail navigation with OAuth model fetching
• Replaced kanban with persistent sub-agent sidebar and new app icon
• Gray context menu highlights and paperclip attachment icon
• Canvas element selection flow and prompt chips improvements

New Features

• Edit workspace files in canvas — Workspace files can now be edited inline in the artifact canvas using Monaco Editor with syntax highlighting for 30+ languages, including save (Cmd+S) and cancel (Escape) shortcuts

Bug Fixes

• Package lock synchronization — Fixed stale package-lock.json preventing CI builds after Monaco editor dependencies were added

Improvements

• Secure file write-back — Added workspace:writeFile IPC channel with path-traversal protection using resolved path boundary checks
• Edit state management — Hardened keyboard handlers with refs pattern to prevent stale closures and reset edit mode when switching between artifacts
• Pinned queued messages bar — Added a notification-style bar above the chat input displaying queued messages with inline edit (Enter to save, Esc to cancel), per-item delete on hover, and clear all functionality to prevent messages from getting lost in long conversations

---

mjfoxai

Documentation Site

Shipped the initial documentation site built with Fumadocs, TanStack Start, and Vercel. The site includes:

• Server-Driven UI (SDUI) — deep dive into the architecture for building dynamic, server-controlled interfaces
• Prompt-to-Feature (PTF) — documentation on the natural-language-to-feature pipeline
• Project pages — overview pages for each active project

Changelog

Added this changelog page with infinite scroll, daily MDX entries, and a scaffolding script for fast daily updates.