Tidepane v0.4.0
November 5, 2026
New
- "Show the Brain orb" setting, off by default
- Read the Chat page's chats; optional routine approvals
- + on a workspace's group starts a chat there; groups fold
- Auto mode changes appearance settings; low thinking option
- Pick any installed Mac voice for the brain
- Drag the companion anywhere; the orb stays inside its card
- 3 s undo window for spoken careful commands
- Settings commands with undo
- Browser, chat, automation, auto and account commands
- Layout commands for workspaces, tabs, panes and terminals
- App command registry with list_commands and ui_command
- show_in_app tool opens Chat, Settings and the rest
- Visual cards in replies
- Sandboxed web view for visuals
- Chats learn to draw visuals; setting
- Bundle the visual libraries
- Transcript keeps visuals and their versions
- Visual blocks become their own log events
- Parse visual blocks and their brief
- Chats working at quit carry on after relaunch
- Show compaction, prompt cache and context tokens
- Replies drawn from their Markdown blocks
- Memory folder watcher, setting, open a note
- Remember and ideas go into the memory vault
- Background tasks show, and their end gets its own card
- Ideas as notes with stable numbers; one-time migration
- app_info tool reports the running build
- Working line at the card's foot; card no longer jerks
- Git history, user edits committed first, safe inside another repo
- The vault folder, safe paths, ownership records and Index.md
- Add, refresh, supersede and forget with user-line protection
- Markdown note format with frontmatter and sections
- Links in your messages show as cards
- Tool calls fold into one line with a terminal box
- Offered options become buttons
- Double-tap fn or click the orb to talk; the mic is off otherwise
- The companion lives in the window
- The companion orb and its card
- The companion card runs the fn conversation inside Tidepane
- Companion sits in the sidebar's empty gap above foot blocks
- Every drawn sidebar block is companion list content
- Views report frames for the companion's placement
- Companion moods, word swelling, breath and glow
- Companion placement: sidebar home, freest corner, hysteresis
- Steer a running turn, progress timeline, calmer card
- Working line, turn footer, mode/model/effort controls, context left, arrival motion
- Follow active Claude account
- Add one-click approve-all header toggle
- Add icons to section tabs
- Add Codex account tabs
- Add Claude account tabs
- Schedule account window pings
- Plan and run account pings
- Switch accounts from the chat header
- Switch account without losing conversation
- Move profile controls into the top bar
- Show sent images in transcript
- Paste and send composer images
- Model and session image plumbing
- Add in-chat file reader
- Link referenced files
- Match mockup layout, labels and finish
- Dark Chat palette from the approved mockups
- Official Claude and OpenAI provider marks
- Offer other worktrees
- Label child repositories with branches
- Suggest child repositories for folders
- Small windows, keyboard, VoiceOver; acceptance scenario
- Show, revert or resume after Stop
- Reopening keeps waits and queues without resending
- Approval rows show every lifecycle state
- Pause when the folder goes or its branch changes
- The full approval lifecycle
- First-run cards and the agent capability table
- Offline retries and reconnecting after a dead process
- Check the folder before a chat starts
- Read the existing worktree from git's error
- Inspect folders; add worktree and init
- Check an agent is installed, new enough and handshakes
- Add a Smooth terminal setting for 120 Hz redraws
- On a usage limit, resume when ready or continue on another account
- Worker badges, inspector and stop scopes
- Workers and their questions in the chat's card
- Worker tools for chats; orchestrator chats only delegate
- Chats start background workers; their questions go through the router
- Route worker questions up the chain, one file editor at a time
- Classify worker requests by what they do
- Session tree, limits and stop scopes
- Codex app-server adapter
- Per-chat rail
- Composer for new chats and running ones
- Conversation with living cards
- Chat page with its sidebar
- Switcher row with Second Brain, Code and Chat sections
- Start, send, stop and answer chats from the app
- Store chats beside their sessions
- Project a session's events into turns and living cards
- Record a short summary of each tool call
- Debug window for background sessions
- Run background sessions in the app
- Background session runtime
- Status reducer with one-answer ledger
- Claude stream-json parser and encoder
- Claude headless command line
- Status list, event log and their tables
- Switch a terminal pane's account from its chip
- Settings › Agents › Accounts and workspace override
- Usage and limit status per account
- Per-account folders and launchers
- Account model with defaults and overrides
- Agents can draft automations for review
- Trust the folder when a Claude job is confirmed
- Background scheduler helper
- Trust the job's folder for Claude on confirm
- Persisted approvals and safe recovery
- Automations section, editor and run views
- Run supervisor lifecycle
- App state, helper enable flow and links
- Per-run launch plan for Claude and Codex
- Per-process hook port and fail-closed boundary hook
- Scheduler control socket
- Scheduler engine with slots, catch-up and repo turns
- Run prompt and finish record
- Decode a version's boundaries
- Boundary check enforced per tool call
- Automation tables and store access
- Cron schedule with timezone and DST rules
- Show the role as the row's name and widen the name column
- Events tab, Activity page and honest Updated times
- Record a capped, redacted log of what agents do
- Roles for agents, from the user or the agent that started them
- Make the Second Brain a page with its own left nav
- Make typed output opt-in
- Typed results, error envelope, annotations and a generated API
- Make the Table view look like the mockup
- Clarify agent session table
- Choose the status colours, classic by default
- One list of what waits on you
- One set of state colours, app-wide
- One orchestrator per workspace
- Orchestrator posts its plan; clearer progress
- Orchestrator you choose, over the agent table
- Table view of every agent, by workspace
- One agent may use the whole terminal total
- Agents get 8 terminals, own browser limit
- PHP highlighting; the sidebar shows an editor pane's file icon
- Material Icon Theme file and folder icons
- Stage, commit, discard, switch branch and diffs in Source Control
- Read-only code project pane with explorer, tabs and Source Control
- Find with a find bar, match highlights and next/previous
- Syntax highlighting with an incremental tokenizer, and a mouse pass
- Editor window with a custom text view
- File codec, commands, EOL handling and screen-aware columns
- Cursors, selections and editing commands
- Atomic edit batches and undo history
- Piece-tree text buffer in TermspaceCore
- Replay how the messages between agents went
- Messages between agents on the board, a Messages tab, kept across launches
- Agent-to-agent messages — message_agent, reply_to_agent, wait_for_reply, messages
- Oh My Pi, and New agent starts Claude, Codex, OpenCode or Oh My Pi
- Agents section — usage, program and arguments, environment, MCP servers
- Contrast, interface text size and imported terminal themes
- Cursor shape, a separate light theme, panel animations, agent colours
- A permissions step in the intro
- A live tour of Tidepane in the first-run intro
- Agents outside Tidepane can be resumed, stopped or brought in
- Programs in a terminal can copy to the clipboard (OSC 52)
Fixed
- No blank page back from Code; @ references files
- ui_press and ui_set run on the main thread
- Use this account moves running agents; a new Claude follows the account
- Agent table fits the sheet; companion drag leaves text selection alone
- Companion card is placed for its real height and never runs off the window
- Companion greets, can be talked over, never asks the break question
- Watcher let go during its own event no longer traps
- Seal visuals off WebRTC and media; cards survive scrolling
- Final-review pass — attribution, repo safety, user lines
- Memory list read once per change, not every redraw
- Background strip matches the message box width
- A turn heard while still answering no longer wipes the card
- Companion mic stops when hidden; focus only goes to a visible terminal
- Final review fixes for the companion (mic safety first)
- Typing while the Brain is busy keeps the first answer
- Review round 1 for double-tap talk
- Globe hint when the key is unset; pill says double-tap fn
- Rest orb pauses with its window; card sized before measured; review round 3
- Step down at narrow widths instead of overlapping
- Resting orb breathes at 10 fps; placement rules live in CompanionPlacement
- Typed ask closes the mic; in front the card stays while asking
- Companion always takes sidebar home when it fits
- Reorder tabs with local drag gesture
- Meter shows the limit closest to full; share Main's control
- Enable terminal tab reordering
- Document tabs, keep Main-only sign-in, trim churn
- Harden shared account tabs
- Make parallel test runs reliable
- Prevent Claude ping session artifacts
- Bound account switch shutdown wait
- An image paste does not type the file name
- Respect pasteboard image precedence
- Harden image paste and delivery
- Harden file reader links
- Centre the section switcher on the whole row
- Lay out the folder card's repositories as uniform rows
- Preserve new-chat choices across sections
- Keep folder warnings visible
- Stop section switcher width shifts
- Retain new-chat choices across sections
- Inspector answer to an ended request says so
- Recheck routed targets before approval
- Resume chats after reconnect exhaustion
- Keep reconnect continuation ahead of new messages
- Adopt branch switch before later edits
- Protect prior work when start status fails
- Route inspector approvals through locks
- Preserve routed edits for review
- Bound reconnects after crash output
- Make status announcements polite
- A capped start status unticks every file in Revert
- A later change needs a new review; inspector answers close the chat card
- Revert does not tick files that had your changes before the chat
- Pause the whole tree on a branch change; cheap max-seq query at launch
- Plan 4 leftovers
- Decide Check again from the account's sign-in status
- Keep folder review read-only
- Tighten AgentCheck failure type, version error and Codex handshake
- Close lost questions, limits, MCP-off, ended supervisors
- Git settings and discarding git commands reach the user
- Worker questions show where and how they should
- An open worker question stays answerable on the latest turn
- Orchestrators only delegate; model names checked
- A chat's turn ending keeps its workers' questions open
- A carriage return reaches the user
- Join every continuation before reading the line
- Writes sees flags before subcommands; heredocs reach the user
- Router expires waiting edits, breaks edit cycles, keeps read-only workers read-only
- Classifier refuses a line with a comment
- Classifier joins backslash-newlines before parsing
- Classifier refuses shell functions and endless cds
- Classifier applies every cd to every part
- Only well-known build folders are build output at depth
- Classifier never trusts a cd it cannot prove
- Classifier closes routine bypasses
- Forgetting a chat ends its workers
- Codex turns end only on turn/completed
- Codex adapter keeps what the router needs
- Final review fixes
- A paused chat's question no longer offers Allow and Deny
- Final review fixes
- Final review fixes
- Keep Claude backups per account
- Date agent departures on the real clock
- Reap shells that exit by themselves
- Keep redrawn lines apart after a resize
- Trim the alternate screen when it narrows
- Don't recreate the host of a closed pane
- Drain autoreleased objects after each message
- Send the hook capability from the environment
- Restore Codex 0.165 sandbox flags
- Authenticate automation hooks with the run token
- Honor one-time picker zone
- Isolate private run credentials
- Stop disable on cancel failure
- Treat a missing daemon as no sessions after a grace period
- Bound numeric inputs and check arithmetic
- Resume in the interrupted worktree
- Preserve ambiguous shell denials
- Account for downtime with monotonic budgets
- Retry recovery after unknown daemon listings
- Preserve dry-run retries and confirmation
- Bind approvals to the current request
- POSIX shell for runs when the login shell is not
- Launch runs through the login shell
- Refuse a backslash in the helper plist values
- Confirm the draft the helper created, by id
- Match key names from word starts only
- Drop the caption that repeats the Needs you heading
- Keep the 'not another browser' line in the instructions
- Bound waits and page tool results
- Retain pane ownership across restarts
- Keep last-turn figures as the fallback
- Fetch live account usage
- Keep success ticks green, plain Needs you caption
- Parse a message before its first frame
- Render markdown tables in chat
- Hear the mic in talk mode (echo cancellation on)
- Make faint text readable in all themes
- Diff tab follows its file; "now" for new commits; chips keep their names
- Bounded command readers run at userInitiated
- Read a bounded command's pipes at the caller's priority
- Read git's pipes on threads of their own
- Recover expired and legacy sessions
- Bind queued targets and tab owners
- Scope cancellation and bound reads
- Handle formatted and shadow input
- Retain telemetry and qualify waits
- Validate refs and input outcomes
- Await explicit navigation completion
- Prevent replay and bound telemetry
- End the undo step at a save
- Agent names in messages without spinner marks
- A fresh Claude session is idle at its prompt
- Claude's hooks run by another agent keep that agent's label
- Outside agents are active by their last turn, not the file's time
- The new-tab button sits right after the last tab
- Workspace names keep clear of the cards and the brain