July 1 Release

This release brings a new, optional project memory system, a major trackers upgrade, expanded voice mode (still alpha), plus the latest Claude models and a long list of fixes.

New Features

Project memory & search

• New Nimbalyst Memory extension: indexes your project notes and surfaces relevant facts to the AI and voice agent for grounded answers.
• Global semantic search in Quick Open (Cmd+Shift+O): find any tracker or document by meaning, with an option to include past AI sessions.
• Memory recall now shows a transcript card with the query and matched source documents, click-to-open.

Trackers

• Reference a tracker item from any document or AI chat: type # to pick an item and insert a live chip showing its current status and title. The AI links tracker items as clickable chips too.
• Link tracker items to one another with relationship fields, including automatic "Linked from" backlinks.
• New tracker views: a tag board, saved views (filter and group), and kanban columns that follow each type's custom status order.
• Customize or reset a tracker type's schema from Settings, with a drift warning when it diverges.
• Edit and delete your own tracker comments.
• Share individual plans (and other full-document trackers) with your team — the shared copy keeps its status, lifecycle, and body in sync, including changes made offline.
• Control whether AI agents can use your trackers per project with an "AI Agent Access" toggle.
• Optional "Shared" column in the tracker table shows which items are shared with the team.
• New nim companion CLI for trackers: list, create, update, comment on, archive, and import items from the terminal.

Voice mode

• Start a new coding session by voice — say "create a new session" on desktop or mobile.
• On mobile, find sessions by topic, switch sessions, summarize a session, answer a session's pending question by voice, and send coding tasks to your desktop.
• The mobile floating mic shows what the voice agent is doing, with clear Pause and Cancel buttons.
• Choose the voice model and reasoning level in Voice Mode settings.

Models & platforms

• Claude Sonnet 5 and Claude Fable 5 are now selectable across the Claude chat, Agent, and Code CLI providers.
• New Android app with email/magic-link sign-in, push notifications, deep links, and a pairing QR scanner.
• New Gemini (Antigravity) extension, usable as an AI chat and meta-agent provider.
• New RTL Support extension: auto-detects right-to-left languages and renders the transcript and input correctly.

Other

• Custom completion sounds — pick your own audio file to play when an agent finishes a turn.
• /session-cleanup command tidies your Sessions board with phase corrections and archive suggestions.
• Dart syntax highlighting in the Monaco editor.

Improvements

• Claude Code CLI sessions store and sync far less redundant data, and defer MCP tool schema loading to cut baseline context usage.
• New AI sessions appear immediately instead of waiting for sync to connect.
• Contextual tips now fill empty AI sessions immediately.
• Updating a tracker item no longer links it to the current AI session unless you ask.
• Linked sessions now appear at the top of a tracker item's detail.

Fixed

Trackers

• Linking tracker items now reliably updates both sides and no longer goes stale or drops other links after syncing, including when the AI sets the link.
• Tracker relationship fields no longer get cleared or dropped by concurrent syncs.
• Tracker status changes now work for custom types that rename their workflow status field.
• Tracker reference links (nimbalyst:// chips) in chat no longer render blank.
• Reopened secondary projects now scope the tracker list to the correct project.
• Tracker type counts no longer briefly flash "0" while data is still loading.
• Fixed tracker field corruption on the SQLite backend caused by merging JSON updates.

AI & sessions

• Claude Code background sub-agents are no longer killed when the lead agent's turn ends.
• AI session status no longer stays stuck on "running" in the mobile app after a turn finishes on desktop.
• Another session can read an OpenAI Codex session's last reply through the session-summary tools.
• Windows: Claude Code CLI chat sessions now start reliably, including with multi-line system prompts.
• Extension AI tools (such as OpenSCAD and Replicad) no longer revert recent file edits by saving stale content over them.

Voice

• Voice mode always speaks in your configured preferred language, including on mobile.
• The iOS voice agent reliably speaks its response after a coding agent finishes a task.
• Voice replies no longer speed up, skip, garble, or overlap near the end of longer responses.

Other

• Toggling an extension on/off via an AI agent now actually restarts its backend, and importer crash errors include the real failure reason.
• Git worktrees with branch-style names and a project's own subfolders inherit the parent project's agent permissions instead of re-prompting.
• On Windows, clicking a file link in chat opens the file instead of a blank window.
• Stop prompting to run the Gemini backend at startup; it now starts only when you use Gemini.
• Committing no longer triggers a burst of slow database queries that briefly hitched the app.
• Desktop release builds now bundle the application correctly.