Editable messages, richer chat, and a refreshed settings screen

You can now fix a typo in a message you already sent, paste a link or :emoji: shortcode and have it just work, right-click anywhere in chat for the actions you expect, and find every workspace preference under one unified settings screen. Specialists pick up a deeper VS Code toolset, workflows fill in more node coverage, and a long list of polish lands across menus, toasts, knowledge plots, and the embedded browser.

You can now edit your own messages

Hover one of your own messages and pick Edit message, or right-click it, or just press Cmd/Ctrl + Up with an empty composer to jump straight into editing your most recent message. The bubble swaps to an inline editor, save with Cmd+Enter, cancel with Esc, and an “(edited)” marker sits next to the last line so it’s clear the message changed. Edit history is kept on the server for a future review surface. Editing applies to your own user messages only — replies from specialists stay as they are.

Chat

A handful of long-requested affordances finally land together:

• Smarter links. Paste a URL over a selection and it becomes a markdown link automatically. Click a link in the composer to edit or remove it from a small popover. Hover a link in a sent message to see where it goes. GitHub issue and pull-request URLs render as compact #<number> chips with title, status, and check details on hover.
• Emoji that works the way you expect. Type :rocket: and it expands to the emoji as you go. Type :rock and a compact autocomplete picks the rest. Click the reaction button on any message for a full picker with search, categories, skin tones, and recents.
• Right-click menus. Right-click a message for reactions, reply, copy, translate, and delete. Right-click a channel in the sidebar for the same actions you already had on the kebab — plus a new Copy Channel ID.
• Start a DM from anyone in chat. Click another member’s name or avatar in a message to open their profile card and start a DM. Existing DMs are reused; new ones are created cleanly.
• Browse channels, faster. The browse-channels dialog has a Select mode for joining several channels at once, and a Join all public action for the rest. The sidebar compass icon now shows the count of joinable channels so you know when there’s something to explore. The dialog itself scrolls properly inside its bounds, and you can send messages in bursts without hitting the previous rate limit as easily.

A new specialist participation control per channel

Each channel now has a specialist participation policy: Allowed, Direct only, or Off. Use Direct only to keep a channel quiet unless someone explicitly @mentions a specialist. Use Off to keep specialists out of the channel entirely — join prompts and empty-channel hints stay hidden too. Plain conversation no longer gets silently routed to the last specialist who replied.

Specialists get a fuller VS Code toolset and early memory work

Specialists working in the embedded VS Code can now read diagnostics, open files, run an allowlisted set of commands, capture terminal output, save files, and apply code actions — each gated by its own permission. Open in Editor and diff line numbers now land on the right file and line instead of falling back to the top of the file.

Behind the scenes, memory for specialists picks up new consolidation and knowledge-bridging groundwork, so summaries stay traceable to their sources and connections to your knowledge garden keep their provenance. Translations sent through OpenRouter now route to the right model again.

Workflows

Workflow nodes pick up more parity across the catalog. The node that searches the code knowledge graph now returns a clear “not ready yet” signal you can retry against, GitHub commit nodes treat credentials as optional, and three vector-store nodes are now properly marked as executable. A separate checkpoint folds in progress across the engine, runtime, MCP, knowledge-graph, browser, and specialist slices.

Knowledge & projects

You can now create a Personal knowledge plot without the dialog silently failing, and clicking Create rapid-fire only ever produces one plot. Shared enrichment artifacts (the summaries and search metadata a knowledge garden builds up) are now stored durably with the right access scope, so collaborators in the same channel see the same enriched results without the underlying state ballooning. Project and knowledge plot icons render as solid color chips with auto-contrast glyphs, and every entity gets a stable color from the moment it’s created — even without a custom icon picked.

A single, predictable settings screen

Personal settings and workspace preferences now share one card style. The personal profile picks up editable title and timezone fields. Workspace code-intelligence options — graph coverage, background throttling, intent synthesis, pipeline auto-start, project priority, coverage and sharing, historical index, and path exclusions — each sit in their own named card with a one-line tooltip on the section heading, so you can scan the whole screen at a glance.

Polish & fixes

Destructive actions in menus now read clearly across chat, projects, knowledge, specialists, and member lists. Menu hover and selected-sidebar contrast are stronger in dark mode, so the row you’re pointing at is unmistakable. Toasts with both a description and an action button now lay out cleanly without squeezing the message, and the close button stops painting a filled background in dark mode. The app’s compact 36×36 mark replaces the older winged logo across the workspace rail, specialist builder, and docs. Embedded mini-apps and the in-app browser initialize more reliably when the iframe loads quickly, with cleaner separation between concurrent browser and editor sessions. Feedback reports now include app version, channel, and process memory diagnostics — with personal paths, env, and arguments redacted. And a top-up to your credit balance now refreshes in the app within a few seconds instead of waiting for a sign-out and back in.