AI Settings

The AI Settings screen controls which AI features are available, how much they may be used, what data they are allowed to see, and how long their results are kept. Find it under Admin > AI Configuration > AI Settings. The whole area is hidden on sites that are not licensed for AI.


End User AI Limit

The maximum number of weighted AI tokens that End User features may consume per calendar month. This pool covers the AI features your users touch directly — AI Search, Document Summarizer, Record Summarizer, Draft Reply, the Workflow Explainer and the help assistant. 0 disables the pool. Enter a value to enable it. The number should include the base monthly tokens that come with your site, not sit on top of them.


Send Data to AI

When ON, users may use the features that send your tenant's own data and documents to the AI — summarizing a document, drafting a reply, and similar. When OFF, those features are unavailable.

This governs user content. Configuration definitions you create, such as Record Type and E-form layouts, may still be sent as reference and context for the Configuration AI features regardless of this setting.


Configuration AI Limit

The maximum number of weighted AI tokens that Configuration features may consume per calendar month. This pool covers AI used in the Admin screens — E-form conversion, building Record Types and workflows, merge templates and SOW Builder. 0 disables the pool. There is no separate "allow Configuration AI" switch: a positive limit is what turns these features on, and 0 is what turns them off.

Configuration tokens are billed separately from End User and Automation tokens, so a heavy configuration session can never eat into the allowance your users or your workflows depend on.


Workflow AI Limit

The maximum number of weighted AI tokens that the AI workflow actions may consume per calendar month — AI Invoice Extraction, Data Match, Document Classification, Auto-Annotate, Split Detection, Record Type Detection, Generic Extraction and Generic Split. 0 disables the pool, and this is the default on a new site. While it is 0 those actions cannot be added to a workflow: they still appear in the action list, greyed out, with a note pointing back to this screen. An existing workflow that reaches one of them will fail with a message saying the limit needs to be set.


Automation tokens scale with how many documents you process rather than with how many people use the site, so this pool behaves differently from the other two — a single overnight batch can consume a great deal.


No monthly limit

Ticking No monthly limit removes the ceiling entirely: every token the AI workflow actions consume will be billed, with nothing to stop a large batch or a misconfigured process from running up significant cost.

This is the only unlimited setting in DocMgt, and it is deliberately awkward to arrive at by accident. You confirm it when you tick it, and while it is set a red warning shows what the site has spent so far this month. Untick it to return to a limit — the value resets to 0, so you must enter the new ceiling you want rather than silently inheriting the old one.


End User AI Retention

When the AI does something for a user, DocMgt saves the result so it can be shown again later. These settings control how long each of those saved results is kept. A nightly process deletes anything past its limit. Usage totals and billing history are never affected — only the saved result itself goes.

Each End User AI feature has its own setting, so you can keep the ones people go back to and retire the ones they do not. For each one you choose:

  • Keep for a number of days — the saved result is deleted once it is older than the number of days you set. This is the usual choice.
  • Keep forever — the result is never deleted automatically. Anything you want gone has to be removed by hand.
  • Do not save at all — offered for AI Chat only. Nothing is written, so nobody can look back at what was asked or answered.


Use the Set all to box at the top to apply one choice to every feature at once: pick Keep for and a number of days, or Keep forever, then click Apply. It deliberately does not offer "Do not save at all", because switching a feature's logging off should be a decision you make for that one feature rather than something a sweep does to the others on the way past.


A very short retention on Chat is worth a second thought. If someone reports a problem with an AI answer, support can only look at what is still stored.


Miscellaneous options for AI processing

Default Date Format — How the AI interprets ambiguous dates when extracting data from documents: is 05/06/2025 the 5th of June or the 6th of May?

  • Auto - infer from document lets the AI decide from the document's own context. 
  • Month/Day/Year (US) and Day/Month/Year (European) force the interpretation, which is the better choice when you know the documents come from one region. 

This can be overridden on individual workflow actions.


Enable MCP Server — When ON, AI clients such as Claude can connect to this tenant at /mcp using a real DocMgt user's credentials, and work with records and documents with exactly that user's rights. When OFF the endpoint is not reachable at all. It is OFF by default. If you enable it, create a dedicated user for MCP access with only the rights that access genuinely needs, rather than connecting as an administrator.


A note on how limits are checked

All three limits are checked at the start of each request or action, before any work begins. A request that starts just under its limit is allowed to finish even if it ends slightly over.

The Automation pool stops at 99% of its limit rather than 100%, leaving room for the action that is about to run. On a small limit a single large document can still carry you past the figure you set, so if you must not exceed a number, set the limit a little below it.



NOTE ON (AI): The MCP server lets an external AI assistant act inside your tenant as the authenticating user. It requires MCP to be enabled by an administrator and may incur extra fees. Check with your sales rep for more information. Please verify all results, as AI can make mistakes.