Every other brain tells you what it knows.
Vitrus also tells you what it doesn't.
[1] runbooks/rate-limit[2] decisions/d-007- No database backup policy is documented yet.
Connects to the tools your team already lives in
GitHub
Notion
WhatsApp
Linear
Drive
Gmail
Calendar
Jira
Confluence
GitLab
Discord
Stripe
HubSpot
Asana
Dropbox
Figma
ZoomYour company already knows the answer.
It's just scattered across a dozen tools.
Decisions live in Slack, context in GitHub, docs in Notion, the rest in people's heads. Vitrus pulls it all into one brain you can simply ask — and unlike a black-box assistant, it shows its sources and tells you what it doesn't know yet. No confident guessing. No made-up answers. Just what's true, where it came from, and what's still missing.
vitrus bench gapeval.see the eval ↗Every other brain tells you what it knows.
Vitrus also tells you what it doesn't — and shows you where every answer came from.
- Confident answers — even when they're wrong
- No sources; you just have to trust it
- Silent about what it's missing
- Your knowledge lives on someone else's servers
- Every claim cites its source
- Names what it doesn't know, as a gap
- Deterministic — same answer, every run
- Your brain in your git repo (Apache-2.0)
Ask a question. Get the answer and the gaps.
This runs on a small sample company brain. Ask something it doesn't know and it won't make it up — it tells you. That honesty is the whole point.
From scattered tools to one trusted answer
Connect & forget
Plug in Slack, GitHub, Notion, email and more in a few clicks — then it syncs on its own in the background. No tagging, no curation, no wiki to maintain. Permissions come along, so people only ever see what they're allowed to.
Ask
“Who owns billing?” “How did we fix the outage?” Ask like you'd ask a colleague — no syntax, no digging through tabs.
Trust the answer
Every answer cites its sources and flags what's missing, so you know exactly how much to rely on it.
Connect the tools your company already runs on
Every source funnels into one company brain, continuously and automatically — issues, messages, docs, tickets and calendars become typed, permission-aware nodes with no one tagging or filing anything. Each sync re-captures source permissions, so people only see what they're allowed to. Markdown is canonical; the index is disposable.
Plus a generic Docs adapter and Model Context Protocol (MCP) bridge — point Vitrus at any MCP server and it ingests too. New sources light up with no UI rework.
Already have notes elsewhere? Import them in one command — an Obsidian vault, a Notion export, your ChatGPT/Claude history, or a GBrain migration — straight into the same glass-box brain.
Built to be trusted
Answers, not a pile of links
Get a written answer with citations — not ten tabs to read yourself.
It admits what it doesn't know
No confident guessing. Missing information is shown as a gap, so you can go fill it.
Your data stays yours
Everything lives in plain files you own and can export anytime. No lock-in, ever.
Knows who can see what
Permissions from your tools are enforced on every answer. Private stays private.
Tells you what needs attention
Surfaces stale docs, unresolved incidents and aging gaps before you have to ask.
Works with your AI agents
Plugs into Claude, Cursor and your own agents so they share the same trusted memory — and write decisions back.
Maps your company as a system
Finds operational inefficiencies — unowned services, bottlenecks, bus-factor and redundant tools — with evidence, not guesswork.
Resolves contradictions
When two sources disagree it shows both sides and helps you settle which is current. The loser is marked stale.
It tells you what needs attention — before you ask
Most tools only answer when queried. Vitrus also watches your brain over time and surfaces what's decaying: knowledge gone stale, incidents left unresolved, and gaps that keep aging. It's the same deterministic gap analysis, made temporal — no LLM guessing. Put vitrus watch on a schedule and it becomes a standing radar for your company's memory.
$ vitrus watch # proactive — what needs your attention 🔔 Needs attention (3): [high] unresolved_incident · "DB outage" 9d [med] stale_knowledge · "Billing policy" 142d [low] aging_gap · "Who owns on-call?" 21d vitrus dream # nightly consolidation (cron)
It maps your company as a system — and shows where it breaks
Consultants spend weeks drawing how your org actually works — and the map goes stale the day they leave. Vitrus reads the same graph of people, services and dependencies continuously, then surfaces the inefficiencies: services with no owner, single-person bus-factor risk, overloaded bottlenecks, handoffs built on superseded ground, and redundant tools. Every finding cites the real nodes behind it — evidence, not a consultant's guess.
$ vitrus ops # the company as a system, severity-ranked [high] broken_handoff · payments → superseded legacy-auth [high] unowned · "gateway" has no owner [high] bottleneck · alice — 8 things depend on her [med] bus_factor · payments — single owner [med] redundant_tool · datadog ⇄ new-relic (0.94)
Two sources say different things? Vitrus tells you — and helps you settle it.
Other brains pick one answer and hope. Vitrus detects contradictions deterministically and shows you both sides — an explicit conflict, or one person listed at two companies. Choose the winner and the loser is superseded (marked stale); the conflict closes and every agent sees the resolution. Nothing is overwritten in silence — and nothing is guessed.
A memory your agents read before they act — and write to after they decide
Vitrus speaks MCP — point Claude Code, Codex or Cursor at your brain and every agent shares the same sourced, permission-aware memory your team uses. They read before they act (search, think, the gap report) and write after they decide — so the brain stays live without anyone writing docs. One command wires it up: vitrus hooks install. And if a new decision contradicts an existing one, Vitrus tells the agent back — it never silently overwrites.
$ claude mcp add --transport http vitrus \ https://cloud.vitrus.dev/t/acme/mcp \ --header "Authorization: Bearer •••" # now your agent can: think → sourced answers + gaps verify → grounded / stale / contradicted gap_report → what's missing, deterministically
Answers in Slack — with the gap-box, not just a guess
Non-technical teammates don't need the dashboard or a CLI. Type /vitrus what did we decide about billing? in any channel and Vitrus replies right there — the sourced answer, its citations, and the same yellow “what the brain doesn't know” box. Other memory tools drop plain recall into chat; we bring the whole glass box. Answers are fail-closed — the bot only ever sees what its configured access allows, so private channels never leak.
And it doesn't wait to be asked. Put it on a schedule and Vitrus posts what's decaying — stale knowledge, unresolved incidents, aging gaps — straight into your channel. The same deterministic gap-radar, no LLM guessing, delivered where your team already is.
Your newest hire's agent works with your senior team's judgment
A new hire — or their agent — doesn't start from zero. Vitrus turns your company brain into a sourced, role-specific learning path: what to read, who to ask, and what isn't written down yet. Every step links to its source. And because it's a skill system on an auditable, open-core brain — not a black box — you can see exactly what judgment it's working from.
Why not just search?
| Search | ChatGPT | Vitrus | |
|---|---|---|---|
| A real answer, not pages | ✗ | ✓ | ✓ |
| Cites its sources | partial | weak | ✓ |
| Tells you what it doesn’t know | ✗ | ✗ | ✓ |
| Proactive — warns you unasked | ✗ | ✗ | ✓ |
| You own the data | ✗ | ✗ | ✓ |
| Runs on your infrastructure | ✗ | ✗ | ✓ |
| Transparent pricing | — | ✓ | ✓ |
Private stays private. Yours stays yours.
Fail-closed permissions
Source ACLs are re-captured on every sync and enforced on every answer. Removed from a channel? Access revoked on the next sync.
Encrypted vault
Connector tokens are stored AES-256-GCM encrypted, isolated per tenant. Secrets never reach the browser.
Audit log
Every authorized query is recorded — who asked what, what was returned, what was excluded by ACL.
Markdown you own
The source of truth is plain files on your disk. Reset the index, rebuild, same answers — export anytime.
Self-host or air-gap
Runs offline with a deterministic embedder — no API key required to start. Your infra, your rules.
Tenant isolation
Org-scoped engine + row-level security: cross-tenant reads are impossible, verified by tests.
Frequently asked
What exactly is Vitrus?
A company brain. Connect your tools, ask a question in plain language, and get a written answer with its sources — plus an honest list of what your company hasn't documented yet.
How is it different from ChatGPT or search?
It only answers from your own data, cites every claim, and shows gaps instead of guessing. A glass box, not a black box — you always see how much to trust the answer.
Where does my data live?
In plain Markdown files you own. Self-host it on your own infrastructure or use our cloud. Either way there's no lock-in — export everything anytime.
Does it respect permissions?
Yes. Permissions from your connected tools are enforced on every answer, fail-closed. People only ever see what they're allowed to.
Can our AI agents use it too?
Yes — Vitrus speaks MCP, so Claude, Cursor and your own agents share the exact same trusted memory your team uses.
Can our agents write back to the brain?
Yes. Agents read before they act and write after they decide — decisions and reasoning are persisted with their sources, so the brain stays live without anyone writing docs. One command (vitrus hooks install) wires it into Claude Code, Cursor or Codex.
Do we have to tag, file or organize anything?
No. That's the failure mode of every wiki — nobody maintains it. Vitrus captures automatically from your connected tools and keeps itself current in the background. No tagging, no curation, no upkeep.
Can non-technical teammates just ask from Slack or WhatsApp?
Yes — you can ask Vitrus in plain language from chat and get the same sourced answer, with citations and an honest note of what's missing. No new app to learn.
Can it find operational problems, not just answer questions?
Yes. Vitrus reads your company as a system and flags inefficiencies deterministically — unowned services, single-person bus-factor risk, bottlenecks, broken handoffs and redundant tools — each citing the real nodes behind it. And when two sources disagree, it surfaces the conflict and helps you resolve it.
Is it really open source?
The core engine and gap analysis are Apache-2.0 licensed and fully featured. The cloud adds managed connectors, a dashboard and team features.
Start free. Grow when you're ready.
Community
- The full brain + gap analysis
- Works with Claude & Cursor
- Unlimited, forever
Cloud Free
- 1 brain · 1 seat
- 2 connectors · 5k nodes
- Hosted for you + support
Pro
- Seats as you grow · all connectors
- 500k nodes · org MCP endpoint
- Team & ACL · audit log · support
Enterprise
- SSO · on-prem / VPC
- SOC 2 · governance
- No lock-in
Open source, runs on Bun, yours to extend
$ bunx @vitrus/core init --pglite $ vitrus import ./brain $ vitrus think "how was the outage resolved" → a sourced answer [1][2] + what's missing $ claude mcp add vitrus -- bunx @vitrus/mcp
Ask a question.
See the answer — and the gaps.
Set up your company brain in minutes — self-host it free, or start on the managed cloud. Or try the live demo first, no signup.