Changelog
What we shipped and when.
Chat, polished
A big reliability and polish pass on chat sessions from real daily use: nothing gets lost, long threads stay fast, and the composer does what you expect.
- Full thread history always loads — no more hidden older messages on long conversations
- Long threads stay fast: recent messages render instantly with Load earlier on demand
- Session explorer now finds every session across both Claude accounts
- Thread history sorts by recent use, so an active conversation is never buried
- Continue a terminal-run session as a chat and its history comes with it
- Mic dictation in the chat composer (on-device, no API key)
- Text zoom in chats and the orchestrator (Cmd +, Cmd -, Cmd 0)
- Shift+Return for line breaks in the composer
- Chats always open scrolled to the newest message
- Right-click a file path to open it in your browser, Finder, or copy it
- Live usage gauges that update as you work (session, weekly, and Fable)
- Native notifications now show the Orcha icon
- Background agents get a live rail you can dismiss or clear
- The orchestrator can navigate the app for you: take me to that project or session
- Stop a turn with the button or Escape; queue messages while it works
- Terminal mouse-scroll fixed
Chat sessions
The beta arrives: every project is a real Claude Code conversation. Chat tabs bring streaming replies, readable tool cards, your models, and permission modes, with terminals one click away.
- Chat session tabs: a full chat UI over your own Claude Code CLI, per project
- Streaming replies with collapsible thinking sections
- Readable tool cards for Edit, Write, Bash, and Read, with syntax-highlighted results
- Permission modes in the composer: Ask first / Accept edits / Full access
- Model picker per chat: Opus 5, Sonnet 5, Haiku 4.5, plus latest aliases
- Image attachments: paste or drop a screenshot and the model sees it
- Slash-command menu: type / for the full command list
- Message queue and interrupt: send while it works, stop a turn any time
- Per-tab thread history with auto-titles, drafts, and prompt history
- The orchestrator now sees and steers chat sessions across every project
- Reliability: crash-proof process handling, error surfacing, rate-limit and offline awareness
One chat runs everything
The orchestrator becomes the home screen — a persistent chat that sees the live truth of every session and can actually drive them: dispatch, stop, approve, all from one window.
- The Orcha chat is the home screen — one conversation that sees and drives every coding session across all your projects
- Ask what’s running and get a real answer with receipts — live session truth from Claude Code’s own registry, never a stale guess
- Dispatch work from chat; the report comes back to the conversation. Say “let me watch” and it opens in a visible terminal tab
- Stop a running agent or approve its permission prompt right from the chat — no hunting for the terminal
- Automations built by talking — daily tasks and loops that dispatch real agents on a schedule
- Integrations: a curated MCP catalog (GitHub, Linear, Notion, Stripe, Vercel, Supabase, and more)
- Create and register projects from chat — they land exactly like the manual New Project flow
- The Library: Sessions, Backlog, Automations, and Integrations in one place
- Needs You, cleaned up — newest first, and closed sessions clear themselves instead of lingering forever
- Pick your model in the composer (Sonnet by default) — always visible, never hidden
- ⌘⇧J summons Orcha from anywhere; ⌘1–9 switches workspaces
- Click any file path in chat to open it in the right project — no more surprise pop-ups
- In-app update notices: know the moment a new version is out, download from your account
Polish and stability
Round of fixes and stability work as we lived in the app ourselves.
- Session status detection reliability improvements
- Crash fixes under heavy multi-session load
- Smaller UI polish across the sidebar and terminal toolbar
First working build
The first end-to-end build we ran day to day: a native macOS home for Claude Code with embedded terminals and workspaces.
- Workspaces with project tabs and embedded PTY terminals (SwiftTerm)
- Live session status via Claude Code’s native hooks — no screen-scraping
- Needs You — every pending permission and finished session in one place
- All-sessions fuzzy search (⌘⇧S) across your entire Claude Code history
- Backlog quick-capture (⌘⇧K) — type the idea, keep working
- Context Hub — compose session context from summaries, memory, and files
- Background agent dispatch, with optional git worktree isolation
- Resources panel (⌘⇧R) — files and transcripts beside your session
- On-device voice dictation — hold ⌥Space, nothing sent to the cloud
- In-app bug reporting with one-click diagnostics export
Side by side
Two projects, one window: run and compare work across projects without losing your place.
- Side-by-side panes: two projects in one workspace
- Richer session management and history
- Command palette and quick project switching
Terminals that behave
The terminal surface grows up: selection survives streaming output, and you can split your work.
- Text selection no longer clears when Claude keeps writing
- Split terminals, up to three panes
- Per-project dashboards with git status and session counts
The foundation
Development begins: a native macOS app for managing Claude Code, built in Swift and SwiftUI.
- Real PTY terminals via SwiftTerm, multiple tabs per project
- Arc-style workspaces that group your projects
- Automatic discovery of your existing Claude Code projects and history
- Local SQLite persistence, no servers, no account