Your company’s memory. Searchable in plain English.
A private assistant trained on your contracts, SOPs, past projects, meeting notes, and policies. Employees ask in plain English. They get the answer back with the exact document it came from. New hires ramp in days, not quarters.
An Internal Knowledge Assistant from Compass is a private AI assistant trained on your own contracts, SOPs, past projects, meeting notes, and policies — hosted on Canadian infrastructure, respecting your existing permissions, and returning every answer with citations back to the source document. New employees answer 80% of "how do we do X" in their first week. Built for Canadian SMBs in engineering, law, real estate, and studios. Engagements typically run $10,000–$30,000 CAD, scoped to your document volume, integrations, and access controls.
One question. An answer with receipts.
A preview of a knowledge assistant built for a Canadian engineering firm. Same retrieval pattern, your own documents, your own permissions.
A new estimator asks a question in Slack. The assistant picks it up.
Four behaviours on every question.
Whether the question comes through Slack, Teams, a chat widget, or email, the pattern is the same. Read everything, find the source, cite it, admit when the answer is not there.
Ingests everything you have
PDFs, Word docs, SharePoint pages, Notion databases, email threads, Slack messages, your DMS. The assistant indexes what it can read and respects permissions on what it cannot.
Understands your vocabulary
Learns your internal acronyms, project codenames, client nicknames, and SOP language. Ask "what is our standard NDA carve-out for IP?" and it knows what you mean.
Answers with citations
Returns a clear answer plus the exact document and section it pulled from. Click any citation to read the source. No hallucinations, no guesses, no "I think".
Says when it does not know
If the answer is not in your documents, the assistant says so plainly and suggests who in the company might know. Trust beats false confidence.
Same engine. Your institutional memory.
Every build is configured for the documents you have, the questions your team asks, and the systems they already use.
Past-project recall for proposal writing
Ask "what was our scope on the Hamilton bridge inspection in 2023?" and get the exact answer with links to the original SOW, change orders, and post-mortem. New estimators ramp 3x faster.
Precedent and clause library
Ask "show me every termination-for-convenience clause we have drafted in commercial leases" and get a side-by-side list with the original matter files. Associate research time drops dramatically.
Listing and offer history
Ask "what did we sell on Maple Street in the last 18 months?" and get every comparable with price, days-on-market, and offer history. CMA prep becomes a 2-minute conversation.
Front-desk SOP assistant
Front-desk asks "how do I process a class-pack refund mid-month?" and gets the exact policy with the matching workflow in your ops manual. New hires answer 80% of questions in week one.
Hybrid retrieval. Canadian data residency.
The specific tools depend on where your documents live and who needs to see what. The principles below hold on every build. Your documents stay in Canada. Nothing trains a public model.
Knowledge questions, answered plainly.
How is this different from ChatGPT Enterprise or Microsoft Copilot?
ChatGPT and Copilot are generic assistants with limited access to your data. We build a private assistant trained on your specific documents, hosted on your infrastructure. It knows your SOPs (Standard Operating Procedures), your past contracts, your internal vocabulary, your client history. Not "trained on the internet", trained on your business.
Where do my documents live? Is anything sent to OpenAI?
Documents are indexed and stored in a Canadian-residency database (Montreal region by default). The assistant uses an enterprise-tier LLM with zero-retention. Nothing trains a public model. Nothing leaves Canada unless you opt in. PIPEDA (Personal Information Protection and Electronic Documents Act) compliant by default.
Does it cite sources, or do I have to verify everything manually?
Every answer links back to the source document and the specific page or paragraph. You can click any answer and read the original text it came from. No hallucinated facts. If the assistant cannot find an answer in your documents, it says so plainly instead of guessing.
How long does indexing take, and how do new documents get added?
Initial indexing of a few thousand documents takes 2-6 hours. After that, the assistant can watch a folder, an inbox, or a DMS (Document Management System) and ingest new documents automatically as they land. SOP (Standard Operating Procedure) updates, new contracts, meeting notes - all searchable within minutes of arriving.
Can different employees see different things?
Yes. Role-based access controls let you scope what each user can query. Sales sees the playbook, finance sees the contracts, partners see everything. The assistant respects your existing permission structure - if a user could not open the document in your DMS, the assistant will not surface it.
How much does this cost to build?
Every knowledge assistant is custom-scoped. The bill depends on: how many sources (one DMS vs. SharePoint + Drive + email + Slack), document volume in the index, whether access controls need to match an existing permission system, integration depth, and whether you need on-premise hosting. Pilots with one source and one user group are the lightest entry. Real numbers come out of the free Bearings call once we have mapped what your team actually asks and what documents hold the answers.
Your senior staff is paid to do work, not answer the same questions.
The average SMB senior loses 6-9 hours a week answering questions that are documented somewhere nobody can find. The assistant on this page finds them in seconds, cites the source, and lets the senior do the work you actually pay them for.
// Related capabilities
// Built for
Plot Your
Course.
A free 30-minute Bearings call. Two operators talking about your business. No deck. No pitch. Just an honest read on whether AI is the right tool - and if so, where I'd start.