Brain is where every source your team touches ends up — documents, web research, Confluence, chat answers, and manual notes — and it's the single base every assistant, generator, and workflow reads from. Every entry is labeled by origin, written to be read by people and AI alike, versioned automatically, and never becomes searchable knowledge with unreviewed sensitive data still in it.
Every wiki entry carries full revision history. Compare any two versions with a word-level diff, and revert non-destructively — a revert creates a new revision instead of erasing the ones after it, so the audit trail always stays intact.
Link entries with typed relationships — depends on, defines, part of, references, and more — and browse the result as a visual graph. It's how a policy document, the glossary term it defines, and the team that owns it stay connected without duplicating content.
Every entry in Brain carries a visible origin badge. A glossary term pulled from a document, a summary an assistant wrote, and a policy your compliance lead typed by hand all look different at a glance — nobody mistakes an AI extraction for a human sign-off, or the other way around.
Every entry is a genuine formatted document, not a raw blob sitting behind a vector index. A person can read it end to end, edit it, or export it to Word or PDF. The exact same content is semantically embedded so every assistant can ground an answer in it. One piece of content, two readers, nothing to keep in sync.
Define the classes your organisation talks about — people, projects, systems, documents, decisions — and the typed relations between them. The AI proposes a class for every entry, a person accepts it, and search and the graph understand how things connect.
Entries carry an effective date and a lifecycle state. The AI extracts the individual statements of decisions and meeting minutes and spots when a newer fact replaces an older one. It only proposes — a person accepts or rejects each replacement, so Brain Search stops quoting outdated numbers.
A recursive splitter breaks long documents into precisely sized chunks for retrieval. But a chunk that's mostly a pipe-delimited table gets detected automatically and routed through a glossary-extraction prompt instead — so a 40-row terminology table becomes 40 searchable entries, not one blob.
Continue, rephrase, expand, fix grammar, or summarize a section without leaving the entry. Point any of those actions at a Brain-grounded prompt — set per entry or workspace-wide — instead of a blind completion, so what gets written stays specific to your actual content instead of generic filler. Draft edits autosave every couple of seconds; publish when you're happy, or discard and keep the last published version untouched.
Two independent, per-entry toggles cover very different needs. Mark an entry Confidential and only you can find it — hidden from every other member's list, search, and Brain-Search answers, while it stays fully usable by your own AI assistants. Mark it AI-Excluded and it's the opposite: visible to the team if you choose, but its content is never sent to any AI model at all — no embedding, no extraction, no Brain-Search context — for anyone, including you.
Content that touches PII can be gated behind review before it's writable, publishable knowledge — configurable per source and workspace. A Confluence import or document upload that turns up personal data sits in Pending Review until a person clears it, working alongside veilSense's anonymization pipeline that keeps sensitive data out of every LLM call in the first place.
Documents, Confluence, web research, chat answers, and manual notes — indexed automatically, each carrying a visible origin badge and, when required, a PII review gate.
Semantic search finds the right entry by meaning. Every entry is a real document a person can read and edit, and the same content grounds every assistant's answers.
Full revision history, diffing, and non-destructive revert — with export whenever you need the document itself.
One Brain, every source, clearly labeled and fully versioned — safe from unreviewed sensitive data, readable by your team and your AI alike.