Your chats are private to you. The bot answers from what your access tier may see; your own My sources count too.
Overview
What the brain holds right now. Numbers refresh every time you open this tab.
Entities by type
Episodes by source (top 25)
Pages
Read an entity's wiki page as the admin or as any access tier sees it, rebuild it, or open the raw text behind a (src: KEY) citation.
—
Graph
Explore the knowledge graph. Click a node to expand it, double-click to open its page, drag to move, wheel to zoom. The Cypher box is read-only; $group is bound to this group.
Selection
Facts
Cypher query
Skills
Reusable instructions for the bot: how to write an offer, a contract, meeting minutes, a brief. In Chat type / to pick one, or just ask — the bot loads a skill itself when a request matches its description. Import a Claude .skill bundle or a SKILL.md, or write one here; text references inside a bundle (md, json, csv) come along, images and scripts do not.
Skill editor
My sources
Your own knowledge: notes, briefs, contracts, anything you want the brain to know for your questions only. Nobody else — no tier, no admin, no bot answer to anyone else — ever sees these. Ask the bot about them from your own account (Telegram) and it will search them alongside the shared brain.
Paste text
Upload files
txt, md, csv, json, html, pdf (with a text layer), docx — up to 8 MB each.
or browse
My email live search only
Connect your own Gmail so the bot can look things up in it when you ask ("what did the client write me about the invoice?"). Nothing is copied: no archive, no index, no graph. Each question runs a live search in your mailbox, only you can trigger it, and disconnecting deletes the access token.
Sources
Where the brain reads from, and where the bot is allowed to answer. Excluded sources are skipped by every ingest path; "Purge" removes what a source already put into the graph.
Groups register themselves the first time the bot sees a message there, as read-only (it keeps archiving, never answers). Switch a chat on to let the bot answer in it. DMs are governed by the per-user allowlist (Access → Users), not by this list.
Add a chat by hand
Telegram chats & topics
Slack channels
Senders, patterns, chats, topics or channels that every ingest path skips.
Add an exclusion
Add chats
Two ways to bring an existing Telegram chat into the brain. Going forward, the bot captures a chat automatically once it is added to it; these are for the history that predates that.
Upload a chat export
Telegram Desktop → open the chat → ⋮ → Export chat history → format JSON, media off → upload the result.json. Re-uploading the same chat is safe (already-stored messages are skipped).
or browse
Archivist account
A dedicated Telegram account you add to chats you want backfilled. It pulls the full history of any chat it is a member of. Nobody shares a personal account; it only sees chats it was added to.
Connect a mailbox once (read-only; a new window opens for Google sign-in), then decide which categories of mail the brain may see and at what depth.
Ingestion
Live state of the ingest pipeline. Queue counters are per server process; token usage counts since the process started.
Queue
LLM calls (this process)
Last ingested per source
Access
Sort chats and people into access tiers. Enforced live — every answer, fact, raw message and wiki page is filtered to the caller's tier on the server (preview a page per tier in Pages). New chats default to the top, most restricted tier; unknown users get the least access.
Fathom calls, one tier per call. Auto-classify rule: a designer on the call → Management (flagged for review); else a management person → Management; else only C-level people → C-level; nobody from the roster identified → C-level (review). Override any call in the table.
One tier per non-chat collection (Upwork, Drive, Portfolio, team roster…). Unset → top tier (most restricted).
Who may query the bot and at which tier. Identity comes from the platform (Telegram id, Slack id, email) — that is what enforcement trusts. Give a name and at least one identity. Every user can also sign in to this console: admins see everything, others get read-only Pages / Graph at their tier plus their own private sources.
Invite a teammate
The link is single-use and valid 14 days. Send it however you like (Telegram, Slack, email). The teammate opens it and signs in with any Google account or email — that account becomes their login, so you do not need to know their email in advance.
Add or edit a user
The brain's roster — names, titles (authoritative roles), Telegram handle, email aliases. After editing, click Apply to graph to re-ingest so pages reflect the changes.
Add or edit a team member
The ordered tier list. A higher level sees everything at or below it. Changing tiers here does not re-tag existing content by itself.
Add or update a tier
Identities
Keep one node per real person, client or project: review suspected duplicates, declare aliases as standing rules, or merge entities by hand.
Pairs with the same normalised name, a shared alias or handle. Merging keeps the node with more mentions.
A rule merges the two now (if both exist) and applies to everything ingested later.
Declare an identity
Setup
Connect everything the brain needs. Keys are saved in the brain database (the single source of truth) and never shown back in full — only a masked preview and a live check.
Rules
The prompt rules the brain follows. Saving applies to everything ingested from now on; changed page rules need "Rebuild page" (or the build-pages pass) to show on existing pages.
Versions
Set up the brain
A guided walk through everything a new company needs, step by step. Each step says what it is for and what happens without it; fill it in by hand, or ask the AI on the right to explain, draft or do it for you. Come back any time — the list reflects the live state.
Set it up for me — describe your company and the AI configures everything it can
Write freely, in any language: what the company does and for whom, your email domain and console address, who is on the team (names, roles, emails, Telegram @handles) and who should be admin, how you name client chats and channels, which spreadsheets, boards and tools matter, your Telegram bot's username and main group id, shared meeting accounts. The AI fills the workspace settings, creates tiers and users, drafts and saves the company notes, saves the rules as yours, and then lists exactly what only you can do: the keys to paste and the buttons to click. It never asks for secrets in chat.
Ask the AI about this step
Setup helper
Ask in plain words, in any language: who has no tier yet, move these Slack channels to the designer tier, merge two duplicate people, exclude a chat, rewrite a page, invite a teammate. It works through the same console API as the tabs, says what it changed, and asks before anything destructive. Runs as the n8n workflow "Brain Setup Helper" (its executions are the debug log).
API costs
What the brain spends on model APIs, as the providers report it — Anthropic (chat bot, image descriptions, graph extraction) and OpenAI (embeddings, voice transcription). Pulled once a day by n8n; "Refresh now" re-reads the last 7 days. Keys live under Setup → Billing.
By model
By API key
Give each use its own key at the provider (for example "brain-graph" for the graph server and "n8n-bot" for the chat bot) and this table tells you what each part costs.
By day
Updates
The product lives in two Git repositories (the brain server with this console, and the n8n workflows). The host updater checks them every 10 minutes; "Update now" asks it to fetch, apply migrations, rebuild the containers, check the console answers, and sync the workflows into n8n. If the console does not come back, the previous version is restored by itself.
What the update brings
Workflows (n8n)
Runs
Audit
Every change made through this console, newest first.