Enterprise AI search

Answers from your own documents, with the source.

People ask in plain words and get an answer with its sources, drawn only from what each person is allowed to see.

6 to 8 weeks
to live on real traffic
Founder-led
no handoff to juniors
Fixed price
agreed before work starts
What did we agree on data retention in the last security review?
Ask anythingAnswers use only what you can open.SearchingChecking 4 sources against your accessWritingCiting as it goesAnsweredFrom 3 of 4 sources. 1 is hidden because you do not have access.

Customer data is deleted 90 days after an account closes 1. Backups keep it for 35 more days, then roll off 2. The March security review confirmed both and added a yearly deletion check, owned by the platform team 3.

  1. 1Data retention policy, Docs
  2. 2SEC-412 Backup retention window, Tickets
  3. 3Security review notes, March, Chat
Ask nextWho owns the yearly deletion check?Does this cover backups in every region?
Sources checked against your access
1Data retention policyDocs, updated 2 OctDoc
2SEC-412 Backup retention windowTickets, closed 14 MarTicket
3Security review notes, MarchChat, #security-reviewThread
4Board pack, risk registerYou do not have access Drive
The problem

Company knowledge lives in many tools, each with its own search and its own permissions

People ask a colleague instead, and an AI that ignores permissions is worse than none.

Over 25%

of the workweek goes to searching for information.

Sourceatlassian.com

660

apps in use, on average, at companies with 10,000 or more staff.

Sourcezylo.com

99%

of organizations have sensitive data that AI can easily surface.

Sourcevaronis.com

How we fix it

People ask once and get an answer they are allowed to see.

What changesWithout itWith it
FindingAsk a colleague, or search five tools.One question, one answer.
TrustNo source, so nobody checks.Every answer links to its sources.
AccessOne missed permission leaks a file.Answers only from what the asker may see.
How it works

What we build

What changes for your teams. We pick the tools to fit the systems you already run.

Sources checked against your access
1Data retention policyDocs, updated 2 OctDoc
2SEC-412 Backup retention windowTickets, closed 14 MarTicket
3Security review notes, MarchChat, #security-reviewThread
4Board pack, risk registerYou do not have access Drive
What changes

Measured against where you started

Example: Acme Inc

What changes at a company like Acme Inc

Illustrative figures, not a client result. Your own baseline is taken in the first week.

MeasureBeforeAfter
Time to find an answerAbout 20 minutesUnder 1 minute
Answers that show their sourcesNoneEvery one
Files shown to the wrong personUnknownNone in tests
Test questions answered correctlyNot measuredOver 90%

How others built it

Another company built the same kind of system and wrote publicly about how it went.

What you keep

Everything runs in your cloud, and stays yours

You own the config, the tests and the runbooks, so you can swap any part later.

The search running in your cloud, with your permissions kept.

Ask anythingAnswers use only what you can open.SearchingChecking 4 sources against your accessWritingCiting as it goesAnsweredFrom 3 of 4 sources. 1 is hidden because you do not have access.

Customer data is deleted 90 days after an account closes 1. Backups keep it for 35 more days, then roll off 2. The March security review confirmed both and added a yearly deletion check, owned by the platform team 3.

  1. 1Data retention policy, Docs
  2. 2SEC-412 Backup retention window, Tickets
  3. 3Security review notes, March, Chat
Ask nextWho owns the yearly deletion check?Does this cover backups in every region?

The first sources connected and the test questions written.

Sources checked against your access
1Data retention policyDocs, updated 2 OctDoc
2SEC-412 Backup retention windowTickets, closed 14 MarTicket
3Security review notes, MarchChat, #security-reviewThread
4Board pack, risk registerYou do not have access Drive
  • An answer quality report against those questions.

  • A runbook your team owns.

Typical building blocks: A permission-synced index (Onyx connectors and ACL sync, hybrid keyword and vector search, contextual retrieval, a reranker, ColQwen for scans) with Graphiti time-aware memory, behind one MCP server that runs as the user. Golden-question evals gate each release.

Alternatives: Supermemory (fast and young, with self-reported benchmarks, so we run MemoryBench on your data first); Mem0; Glean when buying beats building; Cognee. We do not use GraphRAG or LightRAG as the core.

How the weeks run

Live in 6 to 8 weeks

The work happens in your cloud, with the people who will own it.

  1. Weeks 1 to 2Connect the first sources and write the test questions with your team.
  2. Weeks 3 to 5Build the index with permissions and tune answers against the questions.
  3. Weeks 6 to 8Open it to the first teams and hand over the runbook.
  4. After thatRunning on real traffic

    Answers from company knowledge, with sources.

Before we start

  • Admin access to each source and to one cloud account.
  • A named owner for the search, and one for each source.
  • About 3 hours a week from the search owner, and an hour from each source owner.
Fit

Is this the right first build?

It fits when

  • One question means five tools.
  • A search tool shows what it should not.
  • A team on your side will own it after we leave.

It does not fit when

  • You want a hosted tool picked and resold to you.
  • You want a strategy deck in place of working software.
  • Nobody on your side can own it after we leave.

What’s included and what isn’t

Included

  • Up to 3 sources connected, such as a wiki, a shared drive and a ticket system.
  • The permissions of every connected source kept.
  • Up to 2 teams using it at handover.
  • A test set of up to 100 real questions.

Not included

  • Sources with no API or export.
  • Cleaning up or rewriting your documents.
  • More sources, which your team adds after the handover with the runbook.
Fixed quoteOne price, agreed before work starts

6 to 8 weeks. You get a price range before the call and one fixed number after it. That number is the invoice.

  • If the work takes longer than we planned, we absorb the cost.
  • You pay 50% at the start and 50% at acceptance. Builds of 6 weeks or more pay 40% at the start, 30% at the midpoint and 30% at acceptance.
  • If a build starts within 60 days of your Audit results, the full Audit fee comes off its price.
  • We start three new projects a quarter, so your start date depends on when you book.
Book a demoSee starting prices
FAQ

Questions teams ask before they book

Which search stack do you use?

We pick the one that fits the systems you already run. You own the config, the tests and the runbooks, so you can swap any part later. Most often Onyx connectors with Graphiti memory; Glean when buying beats building.

Can it show someone a file they cannot open?

No. The index keeps each source’s permissions, and answers run as the person asking.

Where does our data go?

It stays in your cloud. The index and the logs run in your account.

How do you know the answers are right?

Your team writes a set of real questions with known answers. Every release must pass them.

How long does it take?

6 to 8 weeks for the first sources and the first teams. More sources are added after the handover, using the runbook.

Enterprise AI search, running in your cloud in 6 to 8 weeks.

Answers from company knowledge, with sources. You own the config, the tests and the runbooks.