Your company already knows the answer. Now your people can find it — and trust it.
One private knowledge base across your files, e-mail, chats, ERP and CRM. Every answer cites its source. Every person, and every AI agent, sees only what it is allowed to see. Your data stays under your control, and Olyra keeps it current, secure and improving as a managed service.
What is Olyra Knowledge Engine? Knowledge Engine is a private, permission-aware knowledge base for mid-market companies. It connects to the places your knowledge already lives — file servers, Microsoft 365 or Google Workspace, your ERP and CRM — reads everything, organizes it against categories you define, and lets people and AI assistants search and ask questions in plain language. Every answer comes with citations to the exact source passage, shows whether that source is verified or out of date, and is limited to what the person asking is permitted to see. AI agents get the same governed knowledge, plus controls that keep them grounded and tell them when to stop. It runs privately, on a box in your building or a server in your own cloud account, and Olyra installs and operates it as a managed service: updates, security patching, monitoring, support and continued improvement. Your data and your keys stay with you.
Your company knows more than it can find
Contracts on a file server. Decisions in mailboxes. Procedures in someone’s head. Customer history split across an ERP, a CRM and Teams. The knowledge exists. Reaching it is the problem.
“Who has the latest version?”
Your best people spend their day answering questions that are already answered somewhere. Every interruption breaks two people’s focus.
Knowledge walks out the door
When an experienced colleague leaves, the reasons behind decisions, the exceptions and the client history leave with them.
Wrong version, wrong answer
A superseded price list or an expired policy gets quoted with full confidence because nothing says it is out of date.
AI you cannot trust
An AI tool without evidence guesses. One given broad access sees whatever your old permissions expose. Nobody can tell you where an answer came from or what the AI was shown.
Permissions nobody understands
Years of shared folders and inherited access mean nobody is sure who can see what — until an AI tool starts surfacing it.
Audit and legal fire drills
“Who had access to this?” and “what exactly do we have on this client?” trigger weeks of manual digging.
From scattered to answerable
Knowledge Engine does the work of a very patient, very well-organized librarian — continuously, and under your rules.
It reads everything, including scans
Documents, mail, chats, spreadsheets, databases and SaaS records are read and indexed. Scanned PDFs and images are read with built-in OCR set up for the languages you work in, so the paper trail is searchable too. Everything is then prepared in advance into a high-speed database in several forms — full-text, meaning, a graph of people and things, and structured records — so questions are answered from what is already built.
It organizes by your categories
Client, department, sensitivity, jurisdiction, document type. You define the categories at setup, your data steward reviews and refines them, and the engine applies them across everything — with fast, reversible corrections.
It answers, cites and stays honest
People and AI assistants ask in plain language. Answers cite the exact passage, flag conflicts between sources, and say “I don’t have enough evidence” instead of guessing.
What your people get
Less time hunting, fewer wrong versions, and answers they can check in one click.
Find by meaning, not keywords
Ask the way you would ask a colleague
- Searches by meaning and by exact keyword together, in English, French, Spanish, German, Portuguese and more
- Filter by source, client, department, document type, date or person
- Every hit opens the original, with the exact passage highlighted
- Ask “what did we know on 31 March?” — the whole knowledge base can be viewed as of any past date
Answers you can verify
Cited, conflict-aware, never bluffing
- Answers are built only from passages the asker may see
- Citations point to document, version and exact location
- Conflicting numbers, dates or revisions between sources are flagged, not hidden
- When the evidence is not there, it says so
Always current
Freshness is part of every result
- Each document is marked verified, stale, superseded or disputed
- Answers that lean on an old document tell you, once and clearly
- Owners are asked to re-check only the content people are actually using — one click from an e-mail
- Tracks documents, never scores people
Connects the dots
Clients, suppliers, people, products, places
- Builds a map of who and what appears where, and how they relate
- Merges spellings of the same client or supplier into one entity
- Shows ERP customers, orders and invoices next to the e-mails and contracts about them
- Surfaces themes across large document collections through graph-based summaries
Your taxonomy, your rules
Built for how your business is organized
- Starter categories proposed from your own documents and industry templates
- Rules show their impact before you switch them on
- Review queue for the unsure cases; anything unclear is held as Unknown and retagged later
- Everything is editable in the console — no tickets, no code
Works where people work
A web console and the AI assistants you already use
- A clean web console for search, ask, review and administration
- Connects Claude, Copilot-style tools and your own automations through the open MCP standard
- Each assistant sees only what the person using it may see
- Every assistant call is logged
How Knowledge Engine makes your AI agents more productive and more accurate
An agent is only as good as what it is given. Most agents today get a search box and a hope. Knowledge Engine gives them cited, current, permission-checked company knowledge — and tells them when to stop.
| When an agent has to… | Without a governed knowledge layer | With Knowledge Engine |
|---|---|---|
| Look something up in the middle of a task | Waits a minute or more while a chat model re-reads documents | Gets the answer from a database prepared in advance, fast enough to use inside a live step |
| Brief itself on a customer or task | Works from whatever a search returns and fills the gaps by guessing | Gets one bounded context pack: the customer card, the ERP records its user may see, approved answers, verified passages, past precedents and warnings — with citations and a receipt |
| Quote a business figure | Lifts a number from an old spreadsheet or an e-mail thread | Asks for a named figure, such as open receivables or orders by month, computed from your ERP records through definitions your steward manages, with the exact query shown as provenance |
| Follow your process | Improvises from a general idea of how companies work | Reads your SOPs and policies as cited, versioned steps with roles, thresholds and approval points |
| Handle a question it cannot answer | Answers anyway, with confidence | Abstains when the evidence is not there and says who owns the topic, so the question reaches the right person |
| Send a draft to a customer | Sends it, and someone finds the wrong price later | Checks the draft sentence by sentence first: supported, contradicted, out of date or unsupported |
| Read an e-mail from outside | Treats it like any other text, instructions included | Sees it labelled as untrusted outside content, so a stranger’s text cannot steer it |
| Learn from last time | Repeats the same mistake for every user | Sees the corrections your staff made to earlier work, once a steward approves them |
Grounded, not guessing
Context with the sources attached
- A bounded briefing per customer or task, sized for the agent, with citations and a receipt
- Named business figures from your ERP records through a governed metric layer, with provenance; figures for very small groups are withheld
- Every passage carries its verification and freshness status
- A receipt records exactly what the agent was shown; replay it later and be told if the source has changed
Knows how you work
Process and policy, cited
- SOPs, policies and procedures become versioned, cited steps with roles, systems, thresholds and approval points
- Divergence findings: when tickets or mail show a process running without its approval step, your stewards are told — documents and counts, never people
- Curated answers approved by a steward take precedence for the questions that matter most
Knows when to stop
Abstain, route, restrict
- Says “insufficient evidence” instead of improvising
- Flags conflicting numbers, dates and revisions between sources
- A “who owns this?” hint routes the question to the owning role or your data steward
- Verified-only mode limits customer-facing agents to verified, in-date, internal sources — anything else goes to a human
Checked before it is sent
Right the first time, better the next
- A grounding check reviews each sentence of a draft against your records
- A wrong price, date, number or revision is shown with both values
- The check runs on the box, with no AI model involved, and never edits or blocks on its own
- Staff corrections become precedents the next agent can see, after a steward approves them
Safe by design
An agent sees only what its user may see
- Per-person permissions are enforced in the database, so no agent can go around them
- Outside content is labelled untrusted, so an agent can refuse to mix it with private data or outbound actions
- Residency rules govern what is handed back to an assistant running outside your building
- Every call goes into the tamper-evident audit trail
Measured, not hoped for
Evidence before and after you deploy
- A readiness report says whether your data can support a given agent, and what to fix first, before you buy it
- Evaluation on a golden set of your own real questions
- Replay past human decisions against the engine to see whether it would have agreed
- Relevance tuning with an impact preview and two-person approval; knowledge-gap analytics turn unanswered questions into a steward’s to-do list
Works with the agents you already have. Any assistant or automation that speaks the open MCP standard — Claude, Copilot-style tools, your own in-house automations — connects under its user’s identity or a service account you approve. Olyra’s pre-built AI agents can draw on the same knowledge layer and controls.
No system makes an agent certain. Knowledge Engine checks against what has been ingested, and it says so when it does not know.
What changes when your business runs on what it knows
The features matter less than the days they give back. Picture these moments in your own company. They are illustrations, not customer stories: the situations Knowledge Engine is built to change.
You stop being the bottleneck
Imagine a Tuesday with no queue at your door. The questions that used to wait for you — why that customer has those terms, which supplier to call, what was agreed last year — are answered from the company’s own record, with the source attached. Your time goes back to the decisions only you can make.
Your key person takes a real holiday
Your most experienced planner is away for three weeks, and nobody phones her. When someone retires or moves on, the files, the mail, the exceptions and the reasons stay findable. People change; the company keeps its memory.
A new hire who contributes in weeks, not months
Imagine a new account manager in her second week, answering a customer about an order from two years ago with the contract clause open in front of her. She asks in plain language, within her permissions, and learns the business from its own record instead of waiting for someone to be free.
The quote goes out right, the first time
The current price list, the agreed terms and the right revision are the ones that get used, because every source shows whether it is verified or superseded. The draft is checked against your records before it is sent. The customer is told the right thing, quickly, and nobody has to call back with a correction.
No more “who knows where that is?”
One question, one cited answer, across the file server, the mailboxes, the ERP and the CRM. The hunt through folders and the message to three colleagues give way to a click that opens the exact passage. Your best people get their concentration back.
Audits and due diligence without the scramble
The auditor, the lawyer or the buyer’s team asks, and the answer is retrieved, not reconstructed: the document, its version history, who could see it and the record of what was produced. The fire drill becomes an orderly request.
AI agents you can hand real work to
An agent that guesses has to be watched. An agent that cites its sources, respects permissions and stops when the evidence is missing can be given more. Lookups are fast enough to use in the middle of a task, and there is a record of what every agent read and did.
Decisions that outlast the people who made them
The reason behind an exception is recorded once, approved by your steward, and found by everyone after that. Ask what the company knew on any past date and see it as it stood. The thinking stays, long after the meeting is forgotten.
A business that gets sharper every month
Every source connected, every correction approved and every gap closed makes the next answer better. Olyra keeps the system current, secure and improving, so what your company knows keeps compounding instead of fading. That is a business you can be proud to run, grow or hand on.
The logic behind the feeling. Fewer wrong answers reach customers, because drafts are checked against your records before they go out. There is less rework, because people and agents start from your process, your figures and your approved answers. Questions that do need a person go to the right owner. And every answer can be traced to its source, which is what lets you rely on it.
Why Knowledge Engine instead of the other option on your list
Most buyers weigh Knowledge Engine against one of five alternatives. Each is a sensible choice for some companies. Here is how they typically compare for a 50–500 person operations business whose knowledge is sensitive and spread across many systems.
| The alternative | Where it typically shines | Where it typically falls short here | What Knowledge Engine does |
|---|---|---|---|
| Copilot bundled with your office suite (e.g. Microsoft 365 Copilot, Gemini) | Already in the tools people use; works across that suite’s own content | Centered on that suite’s content; follows today’s permissions, oversharing included, which is why Microsoft advises an oversharing review before rollout | Puts ERP records, file servers, scans and mail next to Microsoft 365 or Google Workspace; checks permission hygiene before go-live; serves Copilot-style tools too, through MCP |
| Cloud enterprise-search platform (e.g. Glean, Coveo) | Broad connector catalogs and a polished search experience | A shared cloud service: your content is indexed outside your company, and cross-border data transfers bring their own paperwork | Runs privately in your building or your own cloud account, with residency rules that decide what may leave — including answers handed to AI assistants |
| “Chat with your documents” (or ChatGPT or Claude over uploads) | Fast to start on a handful of files | Little or no link to your permissions, ERP or audit trail; answers rarely say whether a source is out of date | Permission-checked, cited answers with freshness on every source, and a record of who asked what |
| Build your own (ChatGPT or Claude over your files, open-source RAG, in-house or with a dev shop) | Free to start; a capable team can show a demo quickly | It cannot tell whether its source is right. A home-built agent sees only the few sources it was wired to, so it cannot know whether the document it found is the current, approved or complete one: no freshness, verification, trust class, conflict detection or permission awareness across sources. It answers confidently from a source that may be wrong or superseded. A demo also usually re-reads documents through a chat model for every question, which can take a minute or more: workable for a monthly report, too slow for daily operations or live agents. Then the same walls: a prepared multi-format database, permissions, sign-in, audit, residency, OCR — and someone accountable when it breaks | Connects to all the places your knowledge lives and records how current and how trusted each source is, so an answer comes with evidence of whether it is right. Your data is prepared in advance for speed, all of that comes together in one system, and Olyra installs, updates, supports and operates it |
| Do nothing (SharePoint or file-share search) | No new project | Keyword search inside one system; ERP, mail and shared drives stay separate, and nothing says which version is current | One cited answer across every source, with the current version marked |
What Knowledge Engine brings together
Individual pieces exist elsewhere. In our market research we did not find another product that combines them in one managed system, built and operated for mid-market companies.
- Built for speed. Your data is prepared in advance into a high-speed database in several formats, so questions are answered from what is already built. Lookups can return in as little as a couple of milliseconds, fast enough for agents to use knowledge in the middle of a task, not only to write the monthly report. Results vary with hardware, data volume and query type.
- Knows whether the source is right. Speed is only half of it. Knowledge Engine connects to all the places your knowledge lives and records how current and how trusted each source is, and it flags sources that disagree. An assistant wired to a few folders cannot tell a current, approved document from a superseded one, so it answers with confidence either way. Here the answer comes with the evidence
- Run for you, under your control. Olyra installs, monitors, patches, updates and supports it as a managed service, on a box in your building or a server in your own cloud account — an outcome, not another platform to staff
- Permissions in the database itself. No screen, integration or AI assistant can go around them
- Residency rules that cover AI answers too. They govern what goes to cloud models and what is handed back to assistants, and a simulator shows what a rule would allow before you switch it on
- Your ERP next to your documents. Ready-made read-only connectors for TOTVS Protheus, SAP Business One, Senior and Microsoft Dynamics 365 Business Central, and connectors for other ERPs built and added for each customer, beside file shares, mail and Microsoft 365 or Google Workspace
- Scans and many languages. OCR for scanned PDFs and images, set up for the languages your documents use
- Agent controls in the box. Verification, freshness, trust class, receipts and a grounding check
- Compliance in the box. Tamper-evident audit, two-person approvals, eDiscovery, data-subject requests and verifiable erasure
- Time travel. Search the whole knowledge base as it stood on any past date
- A boundary around employees. Evaluative questions about individuals are refused — the line works councils and the EU AI Act draw
- No training, no copy. Your data trains no one’s models, and Olyra holds no copy of it
- Keeps getting better. Updates arrive in your maintenance window with automatic rollback if a health check fails, along with security patches, new connectors and regular evaluation reports on your own questions
- Cited answers on day 3. A defined engagement that shows results on your own questions before everything is connected
- Your data stays portable. Everything can be exported at any time, and a factory release returns your data in full whenever you ask
Why Olyra runs it
Olyra is an operations and automation consultancy before it is a software company. Knowledge Engine grew out of our Operational Intelligence work: once leaders have one trusted set of numbers, the next gap is the knowledge behind them — the contracts, e-mails, procedures and decisions that explain the numbers and tell people what to do about them. We build the knowledge base around how operations-heavy companies actually work: an ERP at the center, shared drives that have grown for years, e-mail as the real record of customer and supplier history, and teams that work in more than one language. The taxonomy starts from industry templates and your own documents, and your data steward learns to run it alongside us.
Your knowledge stays yours
Sensitive knowledge is exactly why many companies have held back on AI. Knowledge Engine is designed for that situation.
Nothing leaves without a rule that allows it
Set data-residency rules by where content came from (“everything on the São Paulo file server”) or by what it is (“anything Restricted, anything mentioning this client, anything containing a CPF”). A simulator shows what a rule would allow before you switch it on. If in doubt, the engine blocks — and logs the decision.
People see only what they may see
Permissions are enforced inside the database itself, so no screen, integration or AI assistant can bypass them. Existing file-share and SharePoint restrictions can be carried over.
Everything is on the record
Sign-ins, searches, document views, assistant calls, label changes, exports and support sessions go to a tamper-evident audit trail. Sensitive settings need two people to approve and can be rolled back.
Encrypted, with keys you hold
The knowledge base lives on encrypted storage. Passwords and keys sit in a secrets vault and are never shown in plain text. Recovery keys are handed to you at go-live.
Support by invitation only
No one has standing access to your box. An Olyra engineer requests a session with a reason and a duration; you approve it, can watch it live, can end it, and it is recorded.
Built for the auditor’s questions
Identity and MFA, encryption, logging, change management, backup and third-party access are designed to map to SOC 2 and ISO 27001 expectations, and the engine produces the evidence an auditor asks for.
Control mapping is not certification. Certification of your organization, or of Olyra, is a separate exercise, and legal sign-off remains with your counsel.
Answer the hard questions in minutes, not weeks
- eDiscovery. Open a matter, add custodians, place legal holds, run collections, tag privileged material and produce a Bates-numbered package with a chain-of-custody manifest
- Data-subject requests. Find, export or erase everything held about a person across documents and every personal record, with a signed certificate
- Verifiable erasure. Destroys the encryption key for the evidence, removes derived data and issues an erasure certificate; legal holds block it automatically
- Retention rules. Keep-for periods by label or source; the system proposes what is due and nothing is erased without approval
- Access reviews. See who can see what, and why, for any document or person
- Employee-data boundary. Evaluative questions about an individual employee are refused and small-group results are suppressed, so the tool cannot become a surveillance product
Your data, your control. Our service.
Most companies are not looking for another platform to staff. They want the outcome: knowledge that stays accurate, current and secure, with their data under their own control. Knowledge Engine is delivered that way: a setup, then an ongoing managed service operated by Olyra.
| Yours | Olyra’s ongoing service |
|---|---|
| Your data, the knowledge base built from it and every export, at any time | Setup: specifies, installs and hardens the system, connects your sources, and builds the taxonomy and residency rules with your data steward |
| The recovery keys, handed over at go-live | Runs updates in your maintenance window, with automatic rollback if a health check fails |
| The box in your building, or the server in your own cloud account, where it all runs | Monitors the health of the system and applies security patches, without access to your content |
| The decision to approve every support session and every sensitive change | Supports you under an agreement with published response times, in sessions you approve, can watch and can end |
| Your categories, rules and curated answers: the way your business is organized | Adds new connectors and product improvements as your sources and needs change |
| Your audit trail and the evidence it produces | Delivers evaluation reports on your own questions, so you can see answer quality hold and improve |
| The choice of where AI models run, at addresses you control | Handles failover, scaling and restores: more boxes or sites as you grow, and re-activation on replacement hardware |
| Your third-party accounts and their bills: Microsoft 365, Google, cloud AI, Zapier, Apify | Trains your data steward and stays alongside them as the knowledge base grows |
The box and your data are yours, which helps with compliance, insurance, residency and audit questions. What you subscribe to is the service that keeps the knowledge accurate, current, secure and growing: the software, its updates and the team that operates it. You administer everything the product exposes through the console, and privileged work on the sealed system layer is done by Olyra or an approved technician in a session you approve. Olyra holds no copy of your data, and support engineers see content only with your approval.
A point of trust. Your data can be exported at any time. If you ever decide to stop, a factory release exports it in full and returns the hardware clean.
One product, two ways to run it
Both configurations run the same software. The difference is where the AI models run.
| GPU configuration | Non-GPU configuration | |
|---|---|---|
| AI models run | On the box, on GPUs | On approved cloud AI services, in the region your policy allows |
| Best for | Strict residency, high volumes, sensitive content | Smaller sites, lighter hardware, cloud server deployments |
| Restricted content | Processed locally; nothing leaves | Still searchable by meaning on the box; AI extras such as summaries are redacted, skipped or held — you choose per rule |
| Runs on | A box at your site, or a GPU server | A box at your site, or a server in your own cloud account |
Either way, every AI model is reached at an address you control: your own GPU machine, a cloud service under your own account, or the box itself. Olyra does not host AI models for you, so there is no shared Olyra service in the path of your data.
Starts as one box. Grows as you do.
Add machines when your needs grow, whether for safety, for heavy load or for more offices. Nothing is reinstalled or reloaded.
| Stage | What you add | What it gives you |
|---|---|---|
| Single box | Nothing | Everything on this page on one machine, with encrypted, tested backups |
| Failover pair | A second box | If the first box fails, the second takes over automatically. A small witness keeps the two from disagreeing about which one is in charge |
| Cluster | Three or more boxes at one site | Shares exceptional load and keeps serving if a member is lost |
| Several locations | A replica at each office | One master site; every other location syncs from it and serves its own people locally |
| Disaster-recovery standby | A box at a second site | A ready copy far from the first, brought up if a whole site is lost |
Several offices, one source of truth. A cluster lives in one place. If you want the knowledge base in three locations, one site is the master and the others sync from it. When a replica site has local documents to add, it sends them to the master, the master reads and organizes them, and when it next replicates, the local site receives them back. So a site has its own new data after replication, not before, and everything stays consistent.
A new box is admitted with a join code and a hardware check, with no reinstall. For the first large ingestion, spare office PCs or Macs can be lent as temporary processing nodes to finish sooner. How much each stage adds depends on your hardware, data volume and configuration, and we size it with you.
Connects to where your knowledge actually lives
Six sources done well beat two hundred done badly. And when there is no connector yet, you are never stuck waiting for one.
- File servers and shared drives
- Microsoft 365: SharePoint, OneDrive, Outlook, Teams
- Google Workspace
- ERP records: ready-made read-only connectors for TOTVS Protheus, SAP Business One, Senior and Microsoft Dynamics 365 Business Central; connectors for other ERPs such as Oracle NetSuite, SAP S/4HANA, Epicor, Infor and Sage are built and added for each customer
- CRM: Pipedrive, HubSpot, Salesforce
- Ticketing, code repositories and CSV or database exports
- Any database with an ODBC driver
- Any MCP server you already run
- Zapier and Make automations, and Apify datasets, through your own accounts
- A published connector SDK and a signed webhook inbox for everything else
Where it matters most
Knowledge Engine earns its place where what the company knows is valuable, scattered and hard to lose.
Founders planning an exit
Make the business less dependent on you
- Captures what lives in your head and in your senior people’s inboxes
- Cited answers a buyer’s team can check, under permissions
- Supports a cleaner handover and an orderly data room
Private equity and acquirers
Learn the business fast, and keep it when people leave
- Lowers key-person risk through a founder transition and the first 100 days
- Each company keeps its own box and data, run as a managed service
- Bolt-on acquisitions brought in as sources
Engineering firms
Exact details, the right revision, the reason why
- Every answer cites the exact passage and shows how current it is
- “As of” search shows what a specification said on a given date
- Conflicts flagged; it abstains when the evidence is not there
Manufacturers
Plant know-how that survives retirement
- SOPs, work instructions, quality records and ERP data in one place
- Scanned paper read and searchable
- Fast enough to use on the shift, not only for monthly reports
Regulated and governance-led companies
Answers you can defend, with the record to prove it
- eDiscovery, data-subject requests, retention and access reviews
- Tamper-evident audit trail and approvals for sensitive changes
- Residency rules decide what may leave the box
Teams running Six Sigma, Lean, EOS or ISO-style systems
Processes that are written down, current and findable
- Procedures become versioned, cited steps with roles and approval points
- People and agents find the current version, not a stale copy
- Business metrics sit beside the documents
Knowledge Engine is built for companies whose knowledge is large, spread out and sensitive, typically with roughly 50–500 people and a mix of file shares, Microsoft 365 or Google Workspace, an ERP and a CRM. If you only need a chat assistant over a few hundred documents, a lighter tool will serve you well, and we will say so.
A real answer on day 3, not a progress bar
Buyers of knowledge products lose faith around week six of watching a progress bar. The engagement is built so that never happens.
Install and connect one source
Sign-in is federated with your identity provider and the single richest source is connected first — never “everything at once.”
Your own questions, answered
Ten of your real questions asked and answered in front of you, with citations that open the source.
First AI assistant at work
A starter taxonomy is in place, more sources are connected, and your first assistant does one narrow job end to end.
The first evaluation report
Your golden set of real questions, measured results, your first business metrics and a trained data steward.
Ingestion is real work: your data is read, extracted, classified and built into a high-speed database in several formats, which takes time and AI processing. That is why the engagement starts with one high-value source rather than everything. Remaining sources, classification review, residency rules, roll-out to users and go-live sign-off follow over the next four to six weeks. After go-live the work continues as a managed service: Olyra keeps the system updated, patched, monitored and supported, and regular evaluation reports show how answer quality holds as sources are added.
Knowledge Engine FAQ
How is this different from Microsoft Copilot, Glean or ChatGPT over our documents?
Those tools are built to serve many companies from a shared service, or to work inside one suite. Knowledge Engine is built around one company: it runs privately in your building or your own cloud account, operated by Olyra as a managed service, brings ERP records, file servers, scanned documents and mail together with Microsoft 365 or Google Workspace, enforces permissions in its own database, and applies your residency rules even to what it hands back to an AI assistant. It also tells every assistant how fresh and how trusted each source is. You do not have to choose: Copilot-style tools and ChatGPT or Claude can call Knowledge Engine through MCP.
Why not build this ourselves with ChatGPT, Claude or open-source tools?
You can build a demo quickly, and many teams do. Four things separate a demo from a system you can run daily operations on. Speed: a solution that re-reads documents through a chat model for each question can take a minute or more per answer. That is acceptable for a monthly report that takes a day or two to produce, and too slow for people in the middle of a task or for agents working inside daily operations. Knowledge Engine prepares your data in advance into a high-speed database in several formats, so answers come from what is already built. Knowing whether the answer is right: a home-built agent sees only the few sources it was wired to. It cannot tell whether the document it found is the current, approved or complete one, because it has no freshness, verification, trust class, conflict detection or permission awareness across sources. So it answers confidently from a source that may be wrong or superseded. Knowledge Engine connects to all the places the knowledge lives and records how current and how trusted each source is, so the answer comes with evidence of whether it is right. Depth: far more goes in than search — permissions carried over, sign-in, an audit trail, residency rules, OCR on scans, freshness, agent controls, eDiscovery and erasure, updates and backups. Each is a project of its own. Experience and support: Olyra has built and tuned this before, and operates it for you, so there is an accountable team when something breaks and your people spend their time on your questions, not on the plumbing.
Will it make our AI agents more accurate?
It removes the most common causes of wrong answers: guessing without evidence, leaning on an out-of-date document, quoting a number from the wrong place and following instructions hidden in outside content. Agents get cited, current context; they abstain or route to a person when the evidence is not there; and drafts can be checked against your records before they are sent. The evaluation tools show the effect on your own questions, so you do not have to take it on faith. It cannot make an agent certain, and it checks only what has been ingested.
Does it work with the agents we already have?
Yes, if they speak the open MCP standard, as Claude, Copilot-style tools and many agent platforms do. Each agent works under its user’s permissions or a service account you approve, and every call is logged. Your residency rules decide whether content may go to a cloud model at all; an answer an outside assistant receives has, by definition, left the box, which is why those rules apply to it too. Olyra’s pre-built AI agents use the same interface.
Is Knowledge Engine a cloud service?
Not a shared one. It runs on its own box at your site or a server in your own cloud account, and Olyra installs and operates it as a managed service: updates, security patching, monitoring and support. Olyra stores none of your data. An AI assistant outside your building calls the engine through an endpoint you publish yourself, under a per-agent, per-person access token that you can revoke at any time, or runs on the box.
Do you train AI models on our data?
No. Olyra never sees your data, and the residency rules stop cloud AI providers that train on customer data from receiving it.
Which languages does it support?
Knowledge Engine is built to work in many languages, not one. English, French, Spanish, German and Portuguese are supported, with more added as customers need them. Reading, search and answers work in each, and meaning-based search is designed to match across languages, so a question in one can find sources written in another. Scanned documents are read with OCR set up for the languages your documents use. We confirm coverage of your specific languages, including how well they read from scans, against your own documents before you commit.
What if the box breaks?
Encrypted backups run to a destination you choose, and restores are tested and recorded. Olyra re-activates your license on the replacement hardware and restores from your backups under the support agreement. For stricter uptime needs, add a second box as an automatic failover pair, build a cluster for heavy load, or add a standby at another site; see the scaling section above.
Can it run in several offices?
Yes. One site is the master and each other location holds a replica that syncs from it, so people at every site search locally. Data a replica site has to add goes to the master first, which reads and organizes it and then replicates it back; the local site has it after that replication, not before.
What if Olyra disappeared?
Your data, your keys and the box are yours, so nothing your company knows would be lost. The system keeps ingesting, searching, serving your assistants and exporting. What stops is the part that keeps it valuable over time: updates, security patches, new connectors, support and improvement. That ongoing service is where the value keeps coming from, which is why Knowledge Engine is delivered as a managed service and not as a one-off installation.
Can our IT team see inside it?
Yes, without a shell. The console gives read-only diagnostics, redacted logs and a downloadable diagnostic bundle that contains no content. Logs can be forwarded to your own monitoring tools, and your chosen MSP can be given a bounded operator role.
What does it need from us?
A sponsor such as the COO, CFO or IT lead, and a data steward who can give a few hours a week during the first two months. Read access to the sources you want included, somewhere to put the machine or a cloud account, and a decision on data residency.
Can we see it before we commit?
Yes. We can show you a demonstration box running on sample data, so you can see search, cited answers, permissions and the agent controls working. We do not set up a free trial on your own data: bringing your data in means reading, extracting and classifying it and building a high-speed database in several formats, which takes real time and AI processing. That is why the engagement starts with one high-value source and shows cited answers to your own questions on day 3.
Find out what AI can actually do for your operations.
Start with an AI Operations Audit — a fixed-scope diagnostic that maps your workflows, scores your AI readiness, and hands you a prioritized 120-day roadmap. If we don't find real opportunities, you'll know that too.
Request an AI Operations Audit → Prefer to talk first? Book a 30-minute scoping call.