The wiki lied
Pricing changed in a Slack thread months ago. The wiki still shows the old number, so every agent and every new hire repeats it with total confidence.
Governed knowledge for companies deploying AI
Your agents are confidently wrong about your company. Give them receipts.
BuildStackHQ Brain is the knowledge layer where every answer carries its source, owner, and freshness. One governed canon for your team and your agents, with the same permissions.
- Every answer has a receipt
- Agents over MCP
- No per-seat pricing
Example answerVerified
What does the Growth plan cost, and who approves discounts?
The Growth plan is billed monthly at the approved list price. Discounts above the approved threshold need sign-off from the pricing owner.
Receipt
- owner
- Head of sales, pricing domain
- sources
- 3 events from #sales
- version
- v4, approved by its owner
- freshness
- verifies every 60 days
The wall
Every agent deployment stalls the same way.
The agent does not know the company. Deliberate knowledge rots in wikis nobody maintains, and the decisions that matter live in chat history nobody can search.
Search has no governance
Enterprise search returns a pile of embeddings. No owner, no approval, no way to tell a settled decision from an idea somebody floated once.
Trust collapses once
One hallucinated policy in front of a customer and your team stops trusting every AI answer. The deployment stalls and the tooling is quietly abandoned.
How it works
A canon that stays true, not another pile of embeddings.
A small, human-approved canon holds the company's ground truth. A high-volume stream holds everything else. Brain keeps the two honest with each other.
- 01
Cold start drafts your canon
Connect your tools and Brain mines the last ninety days of history into a draft canon: pricing, decisions, processes, org structure. You review a queue instead of writing a wiki.
- 02
Owners approve what is true
Every entry has one named owner, a version history, and an approval record. Approve, edit, or reject in one action. Nothing becomes canon without a human.
- 03
Drift catches contradictions
The stream is compared against the canon continuously. When reality disagrees with an entry, a drafted correction is routed to its owner. Stale entries decay visibly instead of lying quietly.
- 04
Every answer carries a receipt
Answers resolve to source events, an owner, an approval, and a verified date. When the canon has no coverage, the answer says so. No coverage is a correct answer, never a guess.
Five trust labels, one meaning everywhere
The same labels appear in chat, in the review queue, and in every answer an agent receives.
- Verified
- Approved by its owner and inside its verification window.
- Stale
- Approved once, but past its verification date. Shown, never hidden.
- Stream signal
- Found in raw company history, not yet approved into the canon.
- Conflict
- The stream contradicts the canon and the owner has a drafted fix.
- No coverage
- The canon does not cover it, and the answer says so plainly.
For agents
Your agents get the same governed truth your team does.
Governance is the difference between Brain and a wiki or a search index, and it cannot be bolted on afterwards.
MCP native, REST alongside
Any agent framework on any model vendor connects over the Model Context Protocol. The same tools are mirrored as REST for everything else.
Agents are principals
Each agent gets its own scoped key, domain scopes, and rate tier. It queries with the same permissions as the person it acts for.
Every call is audited
Each tool call writes an audit row with the entries it touched and the trust level it served, so you can show exactly what an agent was told.
| Wiki | Enterprise search | Brain | |
|---|---|---|---|
| Where answers come from | Pages someone wrote once | Nearest embeddings, ranked | Approved entries with provenance |
| When facts change | Someone remembers to edit | Old and new rank side by side | Drift detected, owner approves the fix |
| Who is accountable | Nobody | Nobody | A named owner per entry |
| What agents receive | Stale text to hallucinate from | Context with no trust signal | Cited answers with trust labels |
Sources it ingests today
- Slack
- Gmail
- Google Drive
- Outlook
- Notion
- HubSpot
- Fathom
Security
Built to survive your security review.
Permissions are inherited at capture and enforced at query time, never added as a filter afterwards.
Permissions are structural
Access lists are captured with every event and enforced on every query, for people and agents alike. A private channel stays private by construction, not by filter.
The fact crosses, the excerpt does not
A contradiction raised in a private channel still reaches the entry owner, but they only see excerpts they could open in the source tool. Admins do not bypass this.
Tenant isolation in the database
Row level security on every table, asserted by a cross-tenant test suite that blocks merge in CI. Payloads are encrypted at rest.
Everything is auditable
Every approval, rejection, served answer, and erasure writes to an append-only audit log. That is what carries a security review.
Pricing
A platform fee, never per seat.
Human access is free and unlimited, because human engagement keeps the canon alive. Agent queries are metered above an included volume per tier, so your bill grows only as your agents do.
Core
Around$500per month
- Up to 50 employees
- Four connectors
- The full product, no crippled edition
Growth
Around$1,500per month
- Up to 300 employees
- All connectors
- Single sign-on
Scale
From$4,000per month
- Custom retention
- Audit exports
- Dedicated tenancy and compliance reporting
The pilot
30 days, full product, cold start included.
No free tier. Connect your tools and review a canon drafted from your last ninety days. Success is concrete: a reviewed, living canon your team and agents cite by day 30. If the drafted canon is not useful, you get your money back.
FAQ
Questions teams ask first.
Is there a free tier?
No. Cold start is expensive to run, so we offer a paid 30-day pilot instead: the full product with cold start included, and your money back if the drafted canon is not useful.
Do you charge per seat?
Never. Human access is free and unlimited because human engagement keeps the canon alive. You pay a platform fee by company size, and agent queries are metered above an included volume.
What does an agent get when the canon has no answer?
An explicit no coverage answer. An agent saying it has no governed knowledge on something is correct behavior, and often the most important answer it gives.
Can an admin read private messages through Brain?
No. Permissions captured from the source tool are enforced on every read, and the admin role does not override them on raw content.
How do people sign in?
Single sign-on when your company has an identity provider configured, a secure email link otherwise. Only people provisioned in your workspace can sign in.
Small and true beats big and comprehensive.
Start with the facts your agents get wrong today. Brain drafts the canon, your owners approve it, and every answer after that carries its receipt.