Enterprise Search Software for a Regulated Firm: Permission-Aware Answers Across SharePoint and File Shares
A regulated mid-market firm moved off Glean to an Onyx-based private AI stack running inside its own infrastructure, cutting search running cost by 85% while keeping every access rule enforced at query time.
Replacing Glean With a Private Stack the Firm Owns
The firm was already running Glean. It had solved the first problem, which was that knowledge sat scattered across a decade of SharePoint sites, departmental file shares, a wiki two teams maintained and a ticketing system. People could search across all of it, and they did.
What did not work was the shape of the arrangement. The corpus lived in a vendor's cloud, the permission model was a copy rather than the source, and the licence was priced per seat across an organisation where heavy users were always going to be a minority. At renewal those three things were harder to defend than they had been at signature, and the question became whether the same capability could run on infrastructure the firm already operated.
Four Problems Standing in the Way
Per-Seat Licensing Across a Firm Where Use Was Uneven
The Corpus Sat in a Vendor's Cloud
Entitlements Were a Copy, Not the Source
Coverage Stopped Short of the Older Systems
The Solution
We deployed Onyx — the open-source enterprise search and AI assistant platform — inside the firm's own environment, connected to the systems the work actually lives in.
Permission-Aware Retrieval
Onyx syncs each source system's access control lists, so results filter against the individual's real entitlements at query time.
Connectors to Systems in Use
SharePoint, file shares, the wiki and ticketing indexed on a schedule, so answers reflect current documents.
Answers With Citations
RAG over the indexed corpus, every answer citing the sources it drew on — an uncited answer is not usable here.
Models Under Firm Control
Inference runs against models the firm chooses, hosted in its own environment, so no content is sent to an external API.
Rollout by Team
Deployed to one department first, its feedback shaping connector scope before the wider rollout.
Identity & access
Indexing & audit
What It Moved for the Business
One Place to Ask
SharePoint, file shares, the wiki and ticketing are searchable together, so nobody has to guess which system holds the answer.
Entitlements Survived
Access rules are enforced against the source systems at query time rather than duplicated into a model that drifts.
Answers Are Checkable
Every response cites the documents behind it, so a user can open the source and verify before acting.
The Corpus Never Left
Documents, index and inference all remain inside the firm's network boundary, which is what allowed the project to proceed at all.
Adoption That Held
Around 1,000 searches a day across the firm, which is the bar an earlier internal search rollout never reached.
The firm kept its entitlement model intact, kept its corpus inside its own boundary, and gave staff a single place to ask a question in plain language.
Onyx's admin analytics record every query and answer, so the firm now has a running measure of adoption, of what people ask most, and of where the corpus has gaps — the questions that return poor answers are a direct map of the documentation not yet written. If your documents are spread across systems that each search only themselves, and access rules mean you cannot simply index everything into a hosted product, that is the position this build was designed for.
Documents Scattered Across Systems That Only Search Themselves?
Bring a list of the systems you would want indexed and the access rules that govern them. In 30 minutes we'll tell you whether a self-hosted, permission-aware search fits — and what it takes to stand up.