What we have shipped

Connect any remote HTTP or SSE MCP server, or a command-based one that runs inside a sandboxed container. Its tools appear alongside every built-in integration and are permission-gated identically — a third-party server's own claims about how safe its tools are is not trusted for gating.

Every tool an integration exposes now has its own mode: ask every time, always allow, or disabled. Connecting an app no longer grants everything that app can do. A conversation can override a mode for that session only, and the override expires with the chat.

The integrations list moved from a card grid to a single scannable column grouped by category, with connection state, sync health, and per-app permissions on the app's own page rather than a separate global settings screen.

Ask for a companion in plain language in chat and one gets built — name, role, and working style — with a preview to confirm before anything is created. No trip to settings first.

Each companion gets its own file space for what it produces, so it can save a draft mid-task and pick it up later. Companions also gained a model preference, which takes effect on the very next turn.

A task can now fire when new content matches a filter, not just on a clock. A message in a channel you are watching triggers a run, the triggering content is handed to the skill, and the result is delivered unprompted.

An opt-in daily pass that learns your writing voice from your own sent mail, then prioritises incoming email and prepares replies in that voice. It only ever creates drafts — nothing is sent without a person.

Knowledge can now be scoped to a room — narrower than a department — and a Slack channel provisions its own automatically, so anyone who has posted there sees what was shared there. Separately, skills can be given a persistent per-scope container that keeps what they install between sessions.

The Slack bot now resolves the person asking to their Indexbrain account, so personal and team-scoped knowledge is filtered the same way it is on the web. Previously scope filtering was skipped entirely in Slack. Requires reconnecting Slack to grant one new permission.

Every human edit to a knowledge entry is snapshotted, with a full history and a one-click restore that is itself undoable. Entries can also be marked visible only to you — and a preference mentioned to your own coding agent no longer becomes company-wide knowledge by accident.

Labels, drafts, page creation, database schema edits, and calendar events. Every one of them stops for a human approval before it runs, and they are invisible entirely until an admin enables write tools.

Describe a recurring check or report in chat and it becomes a scheduled task — running once, on an interval, or on a calendar, with results delivered to Slack or email.

One visible allowance covering every AI feature, with a usage breakdown by day and by member, and a hard stop rather than a surprise bill. Reading and browsing your own data does not consume credits.

Free, Pro, and Team plans with monthly and annual billing, replacing a fully manual book-a-call flow.

Skills and Workflows were two structurally identical systems, so they became one. Skills are now versioned and immutable once published, scope-authorised, invoked by name in chat, and every run keeps a cited, step-by-step record.