Documents in. Cited answers out.

PG Docs AI is the document intelligence workspace for teams that need receipts. Upload contracts, proposals and policies — ask questions, compare versions, extract structured data and generate reports where every claim links to the page it came from.

Try the demo workspace Watch the 90-second demo

PG Docs AI — Demo Workspace

What are the payment terms in the Acme proposal?

The Acme proposal prices at USD 148,500 for a 12-month term: licensing at $8,000/month, implementation at $34,500 one-off, and support at $18,000/year. Invoices are monthly in arrears with net-30 terms; the implementation fee splits 50/50 between signature and go-live [3]. Late payments accrue 1.0% monthly interest [3]. A 5% second-year discount applies if you commit to 24 months at signature [3].

business-proposal-acme.pdf · p.3business-proposal-acme.pdf · p.4

Searching documents… 3 sections matched

How it works

From document to answer in four steps

A retrieval-augmented pipeline — not a chatbot with a file upload box bolted on.

1 · Upload

Drop in PDFs, contracts, proposals, policies. Processing runs in the background — parsing, chunking with page numbers, embedding.

2 · Ask

Chat with your workspace. Retrieval grounds every answer in your documents — not the model's memory.

3 · Cite

Every claim carries a marker that links to the exact page and source chunk it came from. Verify in one click.

4 · Produce

Summaries, key points, risk registers, structured extraction, side-by-side comparisons and exportable reports.

Citations

Answers you can audit

Every generated artifact carries a numbered source list. Click a marker, read the passage — that's the whole verification loop.

The risk register said

Risk	Severity	Where
Auto-renewal with 90-day notice window [4]	High	p.2
Liability capped at 12 months of fees [6]	Medium	p.6
Service credits are the exclusive remedy [4]	Medium	p.4

The source behind [4]

saas-agreement-globex.pdf · page 2 · chunk 4

“…continues for an initial term of 24 months. It renews automatically for successive 12-month periods unless either party gives written notice of non-renewal at least 90 days before the end of the then-current term.”

The source viewer shows the exact indexed chunks retrieval used — the same text the model saw. Nothing is paraphrased after the fact.

Capabilities

A workspace, not a demo

Everything below ships in the product today — no ‘coming soon’.

Grounded chat

An agentic Claude loop searches before it answers and cites what it used — including saying when the answer isn't in your documents.

Summaries & risk registers

One-click executive summaries, key-point lists and risk registers with severity, location and reasoning.

Extraction templates

Built-in templates for payment terms, parties & dates, obligations and risks — or define your own fields and get structured JSON.

Document comparison

Pick two or more documents; get a difference table, the changes that matter and a recommendation.

Report builder

Describe the report you need, choose the sources, export Markdown with a full source list.

Workspace API

Create API keys per workspace and drive the same pipeline from your own systems — hashed at rest, revocable in one click.

Use cases

Built for document-heavy teams

Procurement teams

Compare vendor proposals line by line, extract payment terms across contracts, and hand legal a cited risk register before signature.

Legal & contract ops

Summarize agreements in seconds, surface one-sided clauses and renewal traps, and answer ‘where is this stated?’ with page numbers.

Compliance & audit

Turn dense policies into checklists, trace every extracted obligation to its source, and export evidence-ready reports.

Research & strategy

Question market reports, synthesize findings across sources, and produce cited briefings instead of unread PDFs.

Security

Isolation by construction

Multi-tenant from the data layer up — see the Security page for the full picture.

Tenant isolation by construction

Every query flows through a tenant-scoped repository that injects the workspace filter — cross-tenant access is structurally impossible, and covered by dedicated tests.

Your keys, your models

Bring your own Claude API key per workspace. Keys are encrypted at rest (Fernet) and write-only through the API — a database leak leaks nothing usable.

Zero-fabrication citations

Citations come from the numbered source list the model was given, not from the model's claims. Markers the model invents are dropped.

Demo

Ninety seconds, end to end

Upload → processing → ask → cited answer → summary → report export.

Demo video placeholder — the recording script and storyboard are in docs/demo-video-script.md.

Scene 1 — upload a contract. Scene 2 — watch it index page by page. Scene 3 — ask about payment terms and get a cited answer. Scene 4 — one-click summary. Scene 5 — generate and export a report.

Pricing

Start free. Upgrade when it sticks.

LLM usage is billed through your own API key — no per-token markup from us.

Free

$0 forever

For trying the full pipeline on your own documents.

1 workspace, up to 3 seats
50 documents
Chat, summaries, extractions, comparison
Bring your own API key
Usage dashboard

Start free

Questions, answered plainly

How do citations work?+

Before answering, the system retrieves the most relevant chunks from your documents and presents them to Claude as a numbered source list (document, page, text). The model must cite those markers inline. Post-processing keeps only markers that actually exist — if the model invents a citation, it is dropped, and if the documents don't contain the answer, the model is instructed to say so.

Which file types are supported?+

PDF, plain text and Markdown, up to 20 MB per document. PDFs are parsed page by page so citations carry real page numbers.

Do I need my own API key?+

On the hosted demo, yes — workspaces run on their own Anthropic key, encrypted at rest, so your usage is billed directly to you with no markup. Self-hosted deployments can also configure server-wide keys as the fallback.

Is my data used for training?+

No. Documents are stored in your workspace's isolated storage, processed only to answer your requests, and never used to train models.

Can I automate this from my own systems?+

Yes. Workspace API keys (Settings → API keys) authenticate the same REST API the app uses — upload, search, extract, report — with the same tenant isolation and rate limits.

What happens if the answer isn't in my documents?+

The assistant says so. The retrieval-augmented prompt explicitly forbids filling gaps with outside knowledge, and our evaluation suite includes negative cases that test exactly this.

Your documents already know the answers. Start asking.

Create a workspace, seed the demo documents, and get your first cited answer in under two minutes.

Create your workspace