Open /brain in TeamX and ask anything. The Brain retrieves from every task, epic, activity, commit, doc, and agent chat your team has — and answers with charts, task cards, code, and citations.
Unlike a generic assistant, the Brain grounds every claim in what actually exists in your TeamX org. It cites what it used so you can click through and verify.
The Brain scans complete_task activities + release notes and replies with a grouped list — by epic, by member, or by area.
Pulls status_breakdown, recent decisions, and the last 3 activities per task. Answers with a chart of where time is going and cards for the blocking tasks.
Searches the decisions surface + spec fields across tasks + the last 20 messages of any chat that referenced it. Cites each source.
Walks activities with touched_files + doc_files under that path. Replies with a file_tree block and a timeline chart.
The Brain tries three paths in order. Whichever hits first answers your question.
When you have Claude Code or Codex connected, the Brain routes your question to your own laptop. Code and prompts stay on your machine. This path is free at inference time because you already pay the model.
Ships in v2 — requires a Brain-bot team_member refactor first. Scaffolding is in place.
If no agent is online but your org has an Anthropic, OpenAI, or Google key in Settings → AI Keys, the Brain calls that model through TeamX’s ai-proxy. Same retrieval, same rich blocks, different wallet.
Live today.
When neither path is available, the Brain still answers — it returns the retrieved items as a Markdown table plus a nudge to connect an agent. Never silence.
Live today. Always available.
The Brain emits structured blocks alongside Markdown prose. Charts render via recharts. Task and epic cards click through to their detail pages. File trees and code blocks render monospaced. Eight block types available today.
chart
bar / line / pie / area — recharts rendered natively
task_card
Click through to the task detail
epic_card
Click through to the epic page
code
Language-tagged code with optional filename
table
Columns + rows with optional title
file_tree
Monospaced list of paths
link_card
External link with title + snippet
text
Standalone Markdown section
The summary index stores short descriptions of each task, epic, activity, doc, and chat — not the full content. Full content is fetched on demand at query time, and only for the 3 most relevant matches per question.
Code contents on your laptop (file source, Edit diffs, shell outputs) never reach the TeamX database. The agent CLI pushes only summarized module-level descriptions when code is part of retrieval.
When the Brain routes through your local agent, your question and the retrieved context go to your machine. The model runs locally or through the account you\u2019ve configured on your agent — TeamX is never the inference provider.
Each member has a private Brain chat. RLS on org_brain_chats means no one else in your team can read your questions or the Brain\u2019s replies.
Every task, epic, activity, agent_conversation, chat_turn, and doc_file in your team. Integration events land as activities so they’re covered too. Code_module summaries are pushed by the agent CLI on your laptop (coming in v2) — file contents never go server-side.
A Postgres trigger enqueues a refresh the moment any source row changes, with a 60-second debounce. A pg_cron job drains the queue every 5 minutes by calling the summarizer. So the index is at most ~5 min behind reality.
No. Each member has their own Brain chat per team. RLS on org_brain_chats enforces self-only reads. Summaries of the org’s data are shared across the team (so everyone’s Brain sees the same org context) but chat messages are private to you.
When routed through the relay, your question and the retrieved org context go to your laptop, where your local agent runs inference. The answer comes back as Brain messages. Nothing sensitive ever hits an external API — your own machine is the endpoint.
You’ll get a short reply saying so, with a hint to connect a local agent or set a BYOK key for fuller answers. The retrieval system is conservative — it scores across keyword overlap, substring match, and trigram similarity, so zero matches usually means the Brain genuinely has no context on the topic.
Not yet. v2 adds an "acting" toggle that lets the Brain call MCP tools (create_task, assign_task, post_slack_message, etc.) after you confirm an action card. v1 is answers-only.
Open /brain in your TeamX dashboard. The empty state has suggested prompts to get you started.