Summarize
What this page is — the capability that turns a finished document into three things a person can actually use: a paragraph for leadership, a plain-language walkthrough, and a table of the commercial terms with the clause each one came from.
What it is for — so the question "what does this contract actually commit us to?" can be answered by someone who was not in the negotiation.
The problem it solves — only the people who negotiated a contract know what it commits you to, and they are not always available.
Route: /org/papers/documents/<document>/ai → Understand → Summarize ·
Permission: Use AI drafting/extraction/Q&A for documents. ·
Utility: papers_summarizer · Cost: 450 credits estimated, 1,800 maximum.
1. What it is
Summarize reads the rendered body, the parties and every filled schema field, and returns three outputs at three different reading levels — not one summary in three formats. Which one you want depends on who is asking:
| Output | Length | Written for |
|---|---|---|
executive_summary | One paragraph, 120 words or fewer | A director who has ninety seconds |
plain_language | Section by section | A non-lawyer who has to operate under the agreement |
key_terms[] | A table | Whoever has to enter the numbers somewhere else |
It reports what the document says. It does not judge whether the terms are good for you, and it is not legal advice:
| Answers | |
|---|---|
| Summarize | "What does this document say?" |
| Risk analysis | "Which parts of it are bad for us?" |
| Extract key terms | "Give me the parties, dates and value as structured data" |
| Fairness / balance | "Whose paper is this, really?" |
Summarize and Extract key terms overlap on purpose but serve different ends: key_terms[] here is
prose-adjacent and citation-bearing, meant to be read; extraction returns typed values meant to be
stored on the document.
2. Why you would use it
The realistic case is not "I have not read this contract." It is "I read it four months ago, three people have asked me about it since, and one of them is on a call now."
- It ends re-reading. The summary is saved to the document, so the second, third and tenth person to ask get the same answer without anyone opening the PDF.
clause_refmakes it checkable. Every key term names the clause it came from, so a summary can be verified in ten seconds rather than trusted blindly. A summary you cannot check is worse than no summary.plain_languageis the handover artefact. The team that has to deliver under an agreement is usually not the team that signed it. This is the version they can work from.- It flags what is missing. The plain-language pass calls out material items the document does not address, which is often more interesting than what it does.
3. What you provide
Nothing. Summarize is a one-press capability with no parameters.
| Input | Required | Notes |
|---|---|---|
| — | — | No parameters. It works on the open document |
The summary is only as good as what is on the page. A document with unfilled fields produces a summary full of gaps — which is accurate, and unhelpful. Fill the fields first, then summarize.
4. What it reads automatically
| Read from the document | Used for |
|---|---|
| Document text — the full rendered body | Everything |
| Title & family — e.g. "MSA", "NDA" | Frames the summary in the right register |
| Parties — everyone named | The "between whom" of the executive summary |
| Key field values — filled schema fields | Amounts and dates, so the summary carries real numbers |
Because it reads the rendered body, a document whose template still carries unresolved merge tokens will be summarised with those gaps visible. That is a genuine signal — see Papers troubleshooting if a summary comes back describing an agreement with no parties or dates.
5. What you get back
| Field | Type | Means |
|---|---|---|
executive_summary | string, ≤ 120 words | What it is, between whom, the money and dates, and the single most important thing to know |
plain_language | string | The document explained section by section in simple language, with missing items called out |
key_terms[] | array | The material commercial terms |
key_terms[].term | string | What the term is — "Payment terms", "Term & renewal" |
key_terms[].value | string | What the document says it is, in the document's own units |
key_terms[].clause_ref | string | The clause it came from, so you can check it |
There are no enums and no severities here — Summarize does not grade anything. If a field comes back empty it is because the document does not say.
6. Worked example
An operations lead is asked, on a call, when a supplier's invoices are actually due. They open the agreement and press Summarize under Understand.
Input: none.
One row of the key_terms[] table returned:
Payment terms value: "Net 45 days from invoice; 1.5%/month late fee" clause_ref: "7. Payment & Invoicing"
What happens next. The lead reads the answer off the table, then opens clause 7 to confirm the wording before quoting it — ten seconds, against the twenty minutes it would take to find the clause cold. The 1.5% monthly late fee was the part nobody had mentioned; it is now the reason the invoice gets paid on time.
The executive_summary from the same run gave the calling director the three sentences they needed,
and the plain_language output was pasted into the delivery team's handover note unchanged.
7. Running it
- Open the document in Orbit Papers.
- Open the AI Assistant drawer, or go to the document's AI workspace.
- Choose Summarize under Understand.
- The result appears in its own tab and is saved to the document.
The result belongs to the document, not to your session — the next person to open it sees this summary rather than paying 450 credits to generate an equivalent one. Use Re-run after the document has materially changed; a re-run is a fresh charge.
Summarize can also be configured to run automatically on submit, on each counterparty round, or
at finalize, from the document type's AI Config tab. Of all the capabilities, this is the one
most worth putting on on_finalize: every finished document then carries its own summary without
anyone remembering to ask. See AI configuration.
8. The admin contract
| Must be true | Where | What happens if it is not |
|---|---|---|
| Your role holds Use AI drafting/extraction/Q&A for documents. | Role editor | The AI Assistant button does not appear |
The papers_summarizer utility is active | Orbit AI Flow → utilities | "This utility is currently disabled" |
| The utility is enabled for your organisation | /org/ai-utilities | Absent from Understand, with no error anywhere |
papers.ai_monthly_credit_cap not yet reached | System Config | "monthly AI credit cap reached: n of n credits used this month" |
| The document has extractable text | The document itself | A scanned PDF with no OCR summarises to nothing. Run OCR first |
For a document marked confidential, auto-run is default-denied — a platform administrator must
set papers.ai_confidential_cloud to the literal value allow, and the block appears only in the
server log. A manual press still works, and where a local model key is configured the run is forced
on-prem and audited on the timeline as ai_local_routed.
9. Don't confuse this with…
| Extract key terms | Returns typed values to store. Summarize returns prose to read |
| Risk analysis | Judges the terms. Summarize only reports them |
| Negotiation summary | Summarises the argument — open points and positions. This summarises the document |
| Ask the repository | Answers across the whole corpus. This is one document |
10. Troubleshooting
| Symptom | Cause |
|---|---|
| The summary is vague and full of blanks | The document's fields are unfilled, or it is an early draft. Complete it and re-run |
| It describes an agreement with no parties or dates | The document's type may carry corrupted merge tokens — the rendered body genuinely has no values in it |
key_terms[] is empty | The document is narrative rather than commercial — there are no terms of the kind it looks for |
clause_ref values look wrong | The document has no clause numbering, so the reference falls back to a heading name |
| The capability is missing from Understand | Role lacks Use AI drafting/extraction/Q&A for documents., or the utility is off for the organisation (§8). The document type does not gate it — the type's AI Config controls auto-run only |
| Nothing ran at finalize despite the auto-run policy | The document is confidential and cloud AI is default-denied (§8) |