§01 · Overview
Find, compare, and explain what matters across your company's documents, with a citation behind every answer.
§02 · Problem
"Where is the document?" is solved. The questions that take time (comparing terms across thousands of contracts, mapping policy drift across subsidiaries, finding outliers in a corpus) require reading the population, not finding the right one.
Today those questions take legal, audit, finance, and compliance teams weeks of manual review. DocPacer produces a cited table in an afternoon.
An example, at population scale
"Across our 5,000 customer agreements, where do liability caps and audit rights diverge from our group standard?"
| Today | Three weeks. A spreadsheet. No way to verify. Nothing audit-ready. |
| DocPacer | A cited comparison table. Every row links back to the exact clause. Auditor-ready by default. |
§03 · Method
DocPacer ingests your document corpus, extracts structured claims from every clause, models them into a queryable graph, runs comparison across the whole population, and links every value back to the exact clause it came from.
PDFs, DOCX, agreements, policies. Any format, any scale, any language.
Extract clauses, parties, dates, amounts, obligations from every document.
Structure findings into a queryable document graph. Every value typed and tagged.
Compare and detect divergence across the whole population. Side-by-side, not one-at-a-time.
Every output links back to the exact clause in the source. The receipt comes with the answer.
§04 · Exhibit A
What the cited output looks like when the question is run against 5,247 customer agreements. Every cell in the citation column links back to the exact clause in the source document. This is the deliverable. Not a chat answer, not a summary.
> Where do liability caps and audit rights diverge from our group standard across all customer agreements?
| Agreement | Liability cap | Audit rights | Status | Citation |
|---|---|---|---|---|
| Acme Corp MSA 2023 | €500K (12× monthly) | Annual · 30-day notice | ✓ | §12.3 · §18.1 |
| FinGroup GmbH SLA | €50K (capped) | None specified | ✕ | §9.2 |
| NordicTech Enterprise | €2M (uncapped) | Quarterly · 14-day notice | △ | §11.1 · §14.4 |
| Meridian Holdings | €500K (12× monthly) | Annual · 30-day notice | ✓ | §10.2 · §17.3 |
| EastBridge Corp | €0 (mutual waiver) | None | ✕ | §8.5 |
§05 · Principles
DocPacer compares entire document populations and cites every answer back to the source clause.
Not a document. Not a folder.
The whole corpus. Thousands of contracts, hundreds of policies, every agreement and obligation across every business unit. Most tools give you one document at a time. A context window is not a corpus. We work at company scale.
Not search. Not summarisation.
Side-by-side divergence detection across the whole set. "Where do these 5,000 contracts disagree with our standard?" is a fundamentally different question from "find me the contract with X." The first is the question that takes a team weeks today.
Not implied. Not paraphrased.
Every cell links to the exact clause in the source document. Usable in audit, litigation, and board reporting. The answer to "how do I know it's right?" is a clickable citation.
§06 · Foundation
DocPacer doesn't compete with the tools experts use. It is the layer underneath them — reading every document once, persisting structure, letting every expert work against the corpus instead of one file at a time.
For one document. DocPacer reads them all.
Chat-with-PDF tools live above us, against a single document at a time. They work for one document or twenty. DocPacer reads the whole population so the same kind of question — "what does this say" — is answerable at corpus scale, with a citation behind every value.
Finds the document. DocPacer reads what's in it.
Search and DocPacer answer different halves of the question — "where is it" and "what does the corpus say." Search sits on top of the index. DocPacer builds the structure the index is pointing into.
New contracts going forward. DocPacer makes the back catalogue legible.
CLM platforms depend on metadata tagged at signing. The 80% of contracts already in your archive are unstructured to them. DocPacer extracts structure retroactively from the corpus they never indexed.
Built by lawyers, for lawyers. DocPacer is built for every expert.
Legal-AI tools live inside one lawyer's workflow. DocPacer is the foundation underneath. The same engine reads the corpus whether the expert on top is legal, finance, audit, procurement, or engineering — each domain brings its own kernels and asks its own questions.
Query-time retrieval. DocPacer does the work upstream.
RAG retrieves plausible chunks at query time. DocPacer does the heavy upstream work — extraction in bounded passes, kernel authoring, citation fidelity, graph persistence — that any retrieval layer needs to land at corpus scale. RAG can live on top of DocPacer. It cannot replace what DocPacer does underneath.
§07 · Use cases
Different roles, different vocabulary, the same kind of question. One that takes weeks today because it requires reading every document. Pick the one that sounds like your week. If yours isn't here, the pattern still applies.
“Across our 5,000 customer agreements, where do liability caps, audit rights, and renewal terms diverge from our group standard?”
Today
A junior lawyer with a spreadsheet, six weeks, and a result the General Counsel does not fully trust.
With DocPacer
A cited table by Friday. Every row links to the exact clause. The work shifts from reading to deciding.
“Across our customer and vendor contracts, where do we carry unlimited liability, off-pattern indemnities, or financial commitments that don't match our group risk appetite?”
Today
An audit-and-rebuild project that ends in a board memo six months later, often after a near-miss has already happened.
With DocPacer
A cited inventory of every clause that creates outsized financial exposure, refreshed whenever you re-run the question — no full re-ingestion. Risk visible before it's realised.
“Across our DPAs, vendor contracts, and subsidiary policies, where do our control claims diverge from the evidence in the documents themselves?”
Today
Sample-pulling, partial reviews, and an ISAE/SOC cycle that depends on the auditor not asking the wrong follow-up.
With DocPacer
A cited control-evidence map across the full population. Gaps between policy and practice surfaced in the same view, every finding traceable to a source clause.
“Across our specifications, runbooks, vendor SLAs, and customer commitments, where do our promises conflict, overlap, or contradict each other?”
Today
Tribal knowledge plus a wiki nobody reads, plus a customer escalation that surfaces the conflict the wrong way.
With DocPacer
A cited consistency map across every spec and contract. Drift detected, not discovered.
“What's standard, what's an outlier, and what should I read first?”
Today
Read for a week and hope. Or pick five at random and call it a sample.
With DocPacer
A population summary in an afternoon. Outliers flagged, citations included. You read what matters and skip what doesn't.
§08 · Pricing
The value lives in the document population we analyse and the depth of analysis we run, not in how many people open a tab.
| Not | Per-employee SaaS pricing. The value lives in the corpus and the depth of analysis, not in how many people have a tab open. |
| But | Priced on the corpus you analyse and the depth of analysis you run. Annual platform access, document population packs, and pooled analysis credits. Use it daily, weekly, or for a one-off question. |
Annual platform fee, document population packs, pooled analysis credits.
One corpus. One question your team cannot answer in a week today. Four to six weeks. Up to 50% credited toward the annual contract if you continue.
§09 · Roadmap · Private beta
Running today against real document populations with design partners. Not GA yet. The full roadmap is its own page, with what's actively building and where it's heading.
Live in design-partner corpora today
Ingest any corpus.
PDF, DOCX, PPTX, XLSX. Long documents, mixed formats, multi-language. The same pipeline whether the population is a hundred contracts or ten thousand.
Cited extraction on every claim.
Every value links to the exact clause it came from. Nothing implied, nothing paraphrased.
Cross-document comparison and divergence.
Side-by-side analysis at corpus scale, not one document at a time. The output is a cited table, not a chat answer.
Queryable through MCP.
Connect Claude, Claude Code, or any MCP-aware client to your corpus and ask questions that span every document. The analyst web UI ships in Next.
§10 · Pilot
Take the customer or vendor agreements from one business unit. Ask where liability, renewal, audit, and termination clauses diverge from your group standard. If we come back with a cited comparison table in an afternoon, you have your pilot.
What we need from you
What you get back
Commercial shape
4–6
Weeks to a cited result
Pricing scoped per engagement. Enquire for a quote.
[ Enquire about a pilot ]§11 · Questions
The questions we get most often in pilot conversations, answered without spin.
RAG retrieves the most similar chunks at query time and asks a model to summarise. DocPacer extracts structured claims up front, persists them as a typed graph, and runs the model in bounded passes with clause-level validation. RAG is good at "find me X." DocPacer is good at "tell me everything that diverges across the whole population." Different shape of question, different architecture.
Every output is anchored to a verbatim quote from a specific clause in a specific document. If the model cannot ground a claim in source text, it returns "not found" rather than a confident wrong answer. Kernels include validation passes that flag low-confidence extractions for human review.
EU-hosted by default. We do not train on customer data. Identity, access governance, and audit-trail tooling deepen through 2026 as pilots formalise their requirements — talk to us about the specifics you need.
Live against design-partner corpora today. The architecture handles populations several orders of magnitude larger than what is currently in production — that scale is not a research bet, it is how the pipeline is built. We will validate at full scale during 2026 against real customer corpora.
Add a new kernel. The corpus is already ingested and graphed. New kernels run against existing nodes without full re-ingestion. New questions return in hours, not weeks.
Yes. The pipeline handles mixed-language corpora. Design partners are running Danish, English, German, Norwegian, and Swedish in the same population.
A kernel is a curated extraction routine that says: for this category of document, extract these specific data points, validate them against these rules, and flag uncertainty. Kernels are how we get from "the model said something plausible" to "the model said something we can stand behind in front of an auditor."
A small founding team with a deep technical bet on document graphs as the right substrate for population-scale analysis. Pilots include direct access to the founders for the duration. For a longer conversation about the technology and the roadmap, write to morten@docpacer.com.