AI capabilities reference
What this page is — every AI capability Orbit Papers offers on a document, in one table, with what each costs, what it needs first, and what it changes.
What it is for — so you can find the right one without opening four groups, and see at a glance which ones will not run on a bare press.
The problem it solves — twenty-three capabilities across four menus, each with its own prerequisites and cost, means people press the wrong one or give up.
Route: /org/papers/documents/<document>/ai · Permission: Use AI drafting/extraction/Q&A for documents.
1. All 23, by group
Needs first marks a capability that will not run on a bare press. Writes marks the two that
change data outside the AI result.
| Group | Capability | What it returns | Needs first | Writes |
|---|---|---|---|---|
| Understand | Summarize | An executive summary, the plain-language version, and the key terms | — | — |
| Understand | Extract key terms | Parties, key dates, obligations, contract value and governing law | — | The obligation register |
| Understand | Missing clauses | Which expected clauses are absent, and which are already present | — | — |
| Understand | Blind spots | What this document omits compared with similar documents in your corpus | — | — |
| Understand | Retention advice | How long this record should be kept, and on what basis | — | — |
| Understand | Classify | The document family, a suggested type, folder and tags | — | — |
| Understand | Entity & concept graph | The parties, assets and obligations, and how they relate | — | The knowledge graph |
| Negotiate | Risk analysis | Risky clauses with severity, rationale and suggested mitigation | Perspective (optional) | — |
| Negotiate | Fairness / balance | How balanced the document is, and which side it favours | Perspective (optional) | — |
| Negotiate | Negotiation summary | Every open point with both positions and a recommendation | — | — |
| Negotiate | Negotiation analytics | Where the negotiation is stalling and who is holding it up | — | — |
| Negotiate | Compromise draft | Draft wording that bridges two stated positions | Two positions — required | — |
| Negotiate | Shadow redline (internal) | Liabilities hidden in wording that reads as harmless — internal only | Perspective (optional) | — |
| Negotiate | Version diff | What actually changed between versions, in meaning rather than characters | A prior version must exist | — |
| Negotiate | Redline grade | Whether the counterparty edit sits inside your playbook | A redline — required | — |
| Comply | Redaction suggestions | Sensitive values that should be redacted before sharing | Audience (optional) | — |
| Comply | Cross-validate | Whether the document agrees with the linked order record | Order reference (optional) | — |
| Comply | Approval routing | The approval path this document should take, and who should review it | — | — |
| Comply | Regional annexures | Region-specific annexures this document is expected to carry | Jurisdictions (only if the document names none) | — |
| Comply | Regulatory drift | Where the wording has fallen behind current regulation | Jurisdiction (optional) | — |
| Comply | Compliance & policy check | Violations of your policy set, with the clause each one sits in | Policy text (optional) | — |
| Comply | Cross-document consistency | Conflicts with the other documents in this contract family | Family relations must be recorded | — |
| Transform | Smart fields | Values for your configured smart fields, with the evidence behind each | Definitions configured in settings | — |
2. What each run costs
Credits are charged per run. A cached result costs nothing to read — only a re-run is charged, and a perspective-aware capability charges separately per perspective.
| Cost (estimated / maximum) | Capabilities |
|---|---|
| 450 / 1,800 | All except those listed below — the standard rate |
| 600 / 2,400 | Cross-document consistency — it reads up to five related documents in one run |
The organisation-wide ceiling is the System Config key papers.ai_monthly_credit_cap. When it is
reached, every capability returns "monthly AI credit cap reached: n of n credits used this month"
until the month rolls over or an administrator raises it. 0 means uncapped.
3. Which can run automatically
A document type's AI Config tab can fire capabilities on three lifecycle events. Nineteen of the 23 can be selected there. Four cannot, and each for its own reason:
| Not available for auto-run | Because |
|---|---|
| Compromise draft | Requires two typed positions. An automatic run has nobody to ask |
| Compliance & policy check | Its value comes from the policy you paste; without one it falls back to generic good practice |
| Fairness / balance | Perspective-dependent, with no perspective to choose from |
| Cross-document consistency | Reads up to five related documents per run, at 600 credits a time |
The other nineteen are selectable on any of the three events. Where each earns its place:
| Event | Fires when | Worth configuring |
|---|---|---|
| On submit | A document is submitted | Approval routing, Cross-validate, Redaction suggestions |
| On counterparty round | Each round arrives | Version diff, Shadow redline, Missing clauses, Negotiation analytics, Redline grade |
| On finalize | A document is finalised | Summarize, Extract key terms, Retention advice, Entity & concept graph |
Redline grade is the interesting one: it normally refuses without a redline,
but on a round event the backend feeds it the newest open redline — which on that event is exactly
what the counterparty just sent back.
Manual runs are always available regardless of what the type configures — the AI Config tab governs automation only, never whether a capability is offered.
See AI configuration.
Auto-run is default-denied for any document whose confidentiality is confidential, and the block
appears only in the server log — nothing on screen says so. A platform administrator must set the
System Config key papers.ai_confidential_cloud to the literal value allow.
Manual runs are not blocked. Where a local model key is configured, a manual run is forced onto
the on-prem model and audited on the document timeline as ai_local_routed.
4. The three that are perspective-aware
Risk analysis, Shadow redline and
Fairness / balance read from a chosen point of view — our_side,
counterparty or neutral — and cache one result per perspective.
Switching perspective is a separate run at a separate charge, not a refresh of the same result. Both answers stay on the document.
5. "Adapt to jurisdiction" is not on this list, and that is not an omission
You may see the name in other material. The product does not offer it. It is grouped with the Comply capabilities in the interface's own configuration, but it is left out of the lists that make a capability appear — deliberately, because it needs a target jurisdiction that nothing in the interface collects. A button for it could only ever fail.
If you need a document rewritten for another jurisdiction, use Regional annexures to find what is expected, and the Drafting AI drawer to write it.
6. Where results go
Every result is stored against the document, so it is visible to anyone who can see the document and survives you closing the page. Two consequences worth knowing:
- A colleague opening the document tomorrow sees your run, rather than paying to repeat it. This is why results are worth generating even when you already know the answer.
- A cached result does not know the world changed. Cross-document consistency caches against this document's content, so a change to a related document does not invalidate it; Regulatory drift is a snapshot of the law on the day it ran. Re-run rather than trusting age.
7. Troubleshooting
| Symptom | Cause |
|---|---|
| A whole group is missing from the rail | Every capability in it is disabled for your organisation. Transform has only one member, so it vanishes entirely when Smart fields is off |
| A capability is missing from a group | Your role lacks Use AI drafting/extraction/Q&A for documents., or that utility is off for your organisation. The document type does not gate it — AI Config controls auto-run only |
| "This utility is currently disabled" | The utility's global is_active switch is off, platform-wide |
| "monthly AI credit cap reached" | papers.ai_monthly_credit_cap is exhausted for the organisation this month (§2) |
| Nothing ran on submit or finalize | Nothing is configured on the type, or the document is confidential and cloud AI is default-denied (§3) |
| "Adapt to jurisdiction" cannot be found | It is not offered (§5) |
| A result looks stale | It is cached with the document. Re-run (§6) |