Tidepane v0.3.12
October 17, 2026
New
- Notes for 0.3.12, the Second Brain
- Board on the left, side panel with chat, tasks and agent
- Composer wraps in a fixed-width box; header and cards polished
- Task board, and a check when you mark an agent's task done
- Task list, shared-worktree warning with handoffs, batch card
- Evidence lens: each answer shows what Tidepane checked, what an agent's log shows, and what the agent only says
- Futures board: each job shows what it waits for and why
- Chat, jobs and the Claude session survive a relaunch
- Personalize: tone presets plus your own words, name, about me
- No message is read before the model; slash commands and set_auto_mode
- Past work is a search_history tool, not a keyword gate
- "tell both" reaches the agents just discussed; one press for several cards
- One ongoing conversation; the model, not word lists, works out what you mean
- One owner for the voice; approving a card stops the answer at once
- Real tool calls instead of text tags, through a per-question MCP server
- "take me there" goes straight to what the chat was about; open reaches plain terminals
- Auto mode runs what you ask without a confirm card; fillers while it thinks
- Memory, watches and multi-step jobs, and hold fn to talk from any app
- The brain looks things up itself: web or code, no mode to pick
- Research and code modes: search the web, or read the agents' folders
- Show what the brain's own Claude calls use; cut each call's overhead
- Cross-agent warnings: one file, two agents; a long wait; a full context
- An ideas inbox: keep an idea in passing, start an agent on it later
- Ask can start an agent, reply to one or open one, on your OK; and answers anything
- Stays on after a restart until turned off
- The brain pulses with its voice, and with yours while it listens
- Talk mode: speak to the brain, it answers out loud and listens again
- Agents connect, let go and glide on the board
- Natural voice with Kokoro, downloaded in the app
- Voice mode, the brain answers out loud
- Readable answers in a chat with the brain
- Write ~/.termspace/brain.json for other tools
- Reply to agents, start new ones, see sessions outside Tidepane
- Agent detail card, Summarize and Ask the brain
- Second Brain page with every agent around one brain
Fixed
- Say plainly when Claude Code is missing or signed out; Sign in button
- Code lookups can't read token files; no user MCP in the fallback; IME Return
- Four gaps: needs-you on started agents, Codex logs, /auto off mid-answer, persona in the backup path
- Auto mode acts only on what you named; safer tags, Return, job waits and timeouts
- Hold fn is on by default, asks for Accessibility once, and logs what fn does
- A conversation carried on in another session is not an outside session
- Ask reads only the agents a question is about; panes keep their transcripts
- No animation on opening or for batches of agents
- Mlx-swift 0.30.6, long sentences no longer buzz
- Own panes' sessions after a relaunch, find far-back prompts