Back office · Documents & Knowledge

Read every document. Surface what matters.

Documents & Knowledge is a system that reads contracts, drawings, RFPs, leases and scanned PDFs, extracts what you care about with page citations, and answers staff questions from your own files. Compass AI Group builds it for Canadian firms with 5 to 50 staff, hosted in Canada and respecting who may see what. Hours of review become minutes of confirming.

About one in six Bearings Reports says: do nothing yet.

Plot Your CourseSee how we work
Every document read, every answer cited
Contracts, leases, drawings and scans, and staff questions, reach the document reader, which produces an abstract with every field cited, answers from your own files cited to the page, writes the record into your systems, and sends low-confidence results to a person for review.Contracts, leasesdrawings, scans, PDFsA staff questionhow do we do thisDocument readerreads, extractsEvery answer links to its pageAbstracteach field citedAnswerscited to the pageHuman reviewlow confidenceYour systemsmatter file or CRMContracts, leases, drawings and scans, and staff questions, reach the document reader, which produces an abstract with every field cited, answers from your own files cited to the page, writes the record into your systems, and sends low-confidence results to a person for review.Contracts, leases, drawings, scansA staff questionDocument readerreads, extractsAbstracteach field citedAnswerscited to the pageHuman reviewlow confidenceYour systemsmatter file or CRMEvery answer links to its page
FIG. 00A sample setup, not a client’s. Documents and staff questions go in; what comes out links back to the page it came from, and anything uncertain goes to a person.

S.01One document

One lease in. A cited abstract out.

The reader goes end to end, pulls every material clause with its page, and surfaces anything non-standard before it lands in your files.

sample-commercial-lease.pdf42 pages · Commercial lease, Ontario
Lease abstract: each field, its value, and the page and section it was read from
FieldValueSource
TenantSample Holdings Inc.p. 1, §1.2
Term10 years, renewablep. 3, §3.1
Base rentPer Schedule B, stepped yearlyp. 5, §5.1
EscalationCPI plus 2% a yearp. 6, §5.4
Renewal optionTwo 5-year renewalsp. 11, §11.2
Personal guaranteeRequired, flagged for reviewp. 18, §18.3

Three items for a person to review

  1. Personal guarantee on a 10-year termNon-standard. Tenants often negotiate this down to a corporate guarantee.
  2. Tenant improvement allowance is vague§7.4 refers to “industry standard” but gives no amount or schedule.
  3. Assignment clause is restrictiveThe usual wording that consent is “not to be unreasonably withheld” is missing. Flag for negotiation.
FIG. 01A sample lease, not a client’s. Invented parties, real mechanics: every field links to its page, and the abstract lands in your matter file for a lawyer to confirm.

S.02What you get

Four behaviours on every document.

The same pipeline whether you feed it a contract, a building plan or a scanned invoice: type-aware extraction, cited answers, and a written record of what was read and decided.

  1. 01

    Reads any document you give it

    PDFs, scans, photos, Word files, drawings. It classifies the type first, then runs the right extraction.

  2. 02

    Extracts with citations

    Every field links back to the page and paragraph it came from. Low-confidence results are flagged for a human pass, not guessed.

  3. 03

    Answers from your own files

    A private assistant that knows your contracts, procedures, past projects and vocabulary. Ask “what was our scope on the Hamilton inspection” and get the answer with the source.

  4. 04

    Writes results into your systems

    Structured data lands in your document system, matter file, CRM or spreadsheet. No re-keying.

Price and start

Scoped in Charting after your Bearings call: fixed scope, fixed price, fixed timeline in writing, before any work starts.

After the build, a Managed Compass plan keeps it running and the index current.Foundation $500Operator $1,200Operator+ $2,500CAD a month

It starts the way every engagement does: Bearings, Charting, Heading, True North (diagnose, scope, build, run). See the Method

S.03One example

Every contract you skim is a clause you missed.

A commercial lease runs to dozens of material clauses. A busy lawyer reads it carefully once and skims it after that. The system reads every clause, every time, on every document, and writes down what it found.

Engineering firm

Past-project recall for the next proposal.

An estimator asks what the firm scoped on a past bridge inspection and gets the answer with links to the original statement of work, the change order and the post-mortem. New estimators can get up to speed in weeks rather than quarters, and the 200-page RFP that arrived this morning is already a requirements list, cited to the page.

S.04Under the hood

Vetted infrastructure. Canadian data residency.

The specific tools depend on your situation. These hold on every build, and you see exactly what is in the stack before scope is signed.

Reasoning
Claude, vision-capable, the current model chosen per build
OCR
Production-grade optical character recognition for scans and photos
Records
Supabase, hosted in Canada (ca-central-1)
Search
Hybrid retrieval: meaning and exact keywords together
Audit
Every field and every answer traceable to its source page
Privacy
Zero-retention model access; nothing trains a public model

S.05Best for

Best for a firm where review work piles up on senior people.

A good fit

  • Reviewing contracts, leases or RFPs by hand eats partner or paralegal hours every week.
  • The answer to “how do we do this” lives in one person’s head or a folder nobody can search.
  • You need every extraction traceable for a regulator, a client or your own peace of mind.

Not yet

  • You handle a few documents a month. Careful reading by a person is still the cheaper answer.
  • Your files are scattered and unnamed. A tidy folder structure comes first; we can plan it in Bearings.
  • You want it to sign off on its own. It reads and flags; a qualified person still decides.

Sometimes the answer is a better process, not AI at all.

S.06Questions

Documents and knowledge, answered plainly.

What kinds of documents can it handle?

Contracts, leases, NDAs, RFPs, technical specs, engineering drawings, financial statements, insurance policies, building plans, court filings, and any PDF or scan. The document type is classified first, then the right extraction runs. Handwritten pages are flagged for a human pass.

Does it cite where it found things?

Every extracted field and every assistant answer links back to the exact page and paragraph in the source. If it says the termination clause is 60 days, you click and see the original text. If the answer is not in your documents, it says so instead of guessing.

Where does our data live? Is anything sent offshore?

Documents are indexed and stored in a Canadian-residency database. Processing runs through an enterprise-tier model with zero retention. Nothing trains a public model and nothing leaves Canada unless you opt in. Built for PIPEDA by default; PHIPA-aware for clinics.

Can different staff see different things?

Yes. The assistant respects your existing permissions: if someone could not open the document in your file system or matter system, it will not surface it. Sales sees the playbook, finance sees the contracts, partners see everything.

S.08Next step

Plot your course.

A free 30-minute Bearings call. You and the Compass team, talking about your business. An honest read on what to fix first, and whether AI belongs in it.

Book Your Bearings Call

Or reach the Compass team directly: batwal@compassaigroup.com, 416-409-2112