AI Terms

1. What the Assistant is

The Assistant is an automated system built on large language models supplied by third-party providers (currently OpenAI’s GPT-6.1 Sol; for a Workspace whose owner has opted in to the Ask AI usage preview or accepted metered usage, Anthropic’s Claude Opus 5.5 instead — in both cases run on our own account at the high reasoning setting unless a User chooses a low, medium or extra-high setting for their own messages; you do not provide a key or choose a model). When you talk to it you are interacting with artificial intelligence, not a human. It can answer questions about the projects you have access to; read the photos and files you attach to a message (the text of a PDF, spreadsheet or document is extracted on our servers and sent to the provider with your message); read the text of the drawings, specifications and documents on your projects and look at their pages as images; help prepare Project Intelligence outputs; work with File Server folders and files; and — at your request — create, update and delete records, add comments, attach photos and files and open pages, using the same permissions your role gives you. It does not do everything a person can do on the screens: measuring, pricing, model conversion and publishing sources stay there.

2. Your content and the results

What you send to the Assistant (“Input”) and what it produces for you (“Output”) are Customer Data under our Terms. As between you and us, you own both. Because of how AI models work, Output may not be unique, and other customers may receive similar output.

3. No training on your content; provider retention

We do not use Input or Output to train AI models, and we do not allow our AI providers to do so. Each provider processes Input and Output to return the response under its business or commercial API terms. For Ask AI on Anthropic we ask for its standard global processing, so Anthropic may process Input and Output in the United States or in other countries; we do not promise US-only processing. OpenAI: we ask it not to store the Assistant’s response records; under its current terms it may still keep Input and Output for up to 30 days to detect and prevent abuse, and some of its features keep data on their own schedules. Anthropic: it keeps API Input and Output for a limited period under its standard retention, and longer where content is flagged under its usage policies. Either may keep content longer where the law requires. We do not have a zero-data-retention arrangement with either provider. Images, including drawing pages the Assistant looks at, may additionally be scanned by a provider for child-safety violations. Our AI providers are listed on our Subprocessors page. We store your conversation history in your Workspace until you delete it.

4. Accuracy and your responsibility to review

The Assistant can make mistakes. It may misunderstand a request, mishear a name, misread a photo or a number, choose the wrong record, or state something that is not true. You are responsible for reviewing Output before relying on it — in particular quantities, hours, prices, dates, safety information, and anything you sign, submit or send to others. Output is not engineering, legal, safety, medical, tax or accounting advice.

5. Actions, approvals and the activity log

Your Workspace administrators choose whether the Assistant must ask for approval before destructive actions (the default), before every change, or never. Changes made by the Assistant are recorded in the activity log and marked as made by the Assistant, under the name of the User who asked. The one exception is the optional Automatic visual review: where it has been turned on for a project, it saves its draft review and that review’s PDF without a separate approval for each report, as described there; every other change follows your administrators’ approval setting. You are responsible for the actions the Assistant takes at your request or with your approval, as if you had taken them yourself.

The Assistant does not sign records, does not send your records to anyone outside your Workspace, and cannot exceed the permissions of the User it is working for.

At your request, and with the approval your administrators require, the Assistant can make an outside read-only link to a File Server item. It never sends the link anywhere: the link is shown to you once in the Assistant, is not kept in the conversation, and you decide whether to share it.

6. Voice, photos and recording other people

Dictation and hands-free voice use your device’s microphone only while you have switched them on, and a mute control is always available. Audio is sent to OpenAI to be transcribed; we do not store it. Spoken replies are read aloud by a voice installed on your device, which sends nothing, or — when we offer natural voices and you choose one — by a natural voice OpenAI generates from the reply text. A natural voice is synthetic: AI-generated, not a recording of a person. Where neither is available, replies are shown as text only. Stop, mute, or talking over a reply ends it. Photos you attach in the Assistant are stored in your Workspace like any other photo.

Recording consent. Laws in many U.S. states require the consent of everyone in a conversation before it is recorded. If other people may be captured when you use voice features, you are responsible for telling them and obtaining any consent the law requires, and for following your own company’s policies and any site rules.

No biometric identification. PMT1 does not use face recognition or voice recognition to identify individuals, and you must not use the Assistant to do so.

7. What you must not use the Assistant for

• Decisions that have legal or similarly significant effects on a person — for example hiring, firing, discipline, pay, or eligibility for work — without meaningful human review.
• Anything that is the sole basis for a safety-critical, structural, electrical, engineering or medical decision.
• Identifying people from their face or voice, inferring sensitive characteristics about them, or monitoring workers covertly.
• Generating fraudulent records, signatures, certifications or safety documentation.
• Trying to extract the underlying models, circumvent safety measures, or develop competing AI models from Output.
• Anything prohibited by our Acceptable Use Policy or by our AI provider’s usage policies.

8. Automatic visual review of drawing revisions (optional)

Automatic visual review is off on every project until a named User who may manage the project’s plans, use Project Intelligence and use Ask AI turns it on for that project, after reading its notice. It is available only where your Workspace owner has opted in to Ask AI on Anthropic. When a new revision of a drawing is then published on that project, the Assistant reviews it beside the revision it replaced, running as that User with their access as it is at each step; it stops if they lose that access or the review is turned off. It sends the drawing pages it is permitted to read (as images and text) and the related project sources that User may open — specifications, addenda and other project documents — to Anthropic, under the same terms as other Ask AI processing described above.

It is paid Ask AI usage at the rates your Workspace has accepted, and it counts against the existing company and per-person budgets and our own spending limit described under Usage, budgets and action packs. Turning it on authorises one thing only: for each reviewed revision, one AI draft review saved with its PDF in the project’s Documents, without a separate approval card for each report. It does not create, approve or send RFIs, bid requests or change orders, or change any other record; a person can make drafts from a review through the ordinary approvals.

If a paid request’s outcome is unknown, it is not sent again automatically; its reservation may stay counted against the budget, and the User it runs as decides whether to retry. A review is an AI draft for a person to check: it can miss differences, a review that could not read everything it needed is marked partial, and nothing in it is a guarantee that every change was found.

9. Precon document processing

Precon can read your bid documents with a dedicated model (Anthropic) to report findings with citations, write briefs, suggest takeoff items and answer questions grounded in those documents. Nothing is read until a Workspace owner approves a quote for that project. To prepare the quote, the full content of the documents you select is transmitted to the provider’s token-count API so the quote is exact; no paid completion is made and nothing is charged for the count. The quote states the documents and versions, the work covered, an itemised estimate, the margin and a hard maximum; approval is recorded with who approved it and when. You then pay the usage actually consumed plus the disclosed margin, never more than the approved maximum; using less costs less. Usage the provider could not confirm is shown as pending, is not billed as known usage and is reconciled by us from the provider’s records. A change of documents, scope, model or price needs a new quote; each approved quote adds to the project’s cumulative maximum. The model’s findings are for your estimator’s review and are not a promise of accuracy; nothing it produces is a construction estimate line or a price.

10. Usage, budgets and action packs

Legacy assistant actions. Until a Workspace owner opts in to the Ask AI usage preview, each Plan includes a monthly allowance of Assistant actions, shown in the Assistant and on our pricing page. One question or one task for the Assistant is one action. The allowance starts again on the first day of each calendar month (UTC); unused included actions do not carry over.

When the allowance is used up the Assistant pauses until the next month, until you change your Plan, or until an owner or admin of your Workspace buys an action pack. A pack is a one-time purchase at the price shown when you buy it. It does not renew and is never bought automatically. Purchased actions are used only after the month’s included actions are gone, and what is left carries over from month to month for as long as your subscription is active. If your subscription ends, unused purchased actions are kept with your Workspace for the 30 days in which you can come back, and are then removed with it. Purchased actions have no cash value and cannot be transferred to another Workspace. Refunds of packs are covered in our Refund & Cancellation Policy.

The Ask AI usage preview. A Workspace owner may opt in, explicitly and early, in Settings → AI usage after reading the price and the provider notice; that opt-in also authorises Ask AI to use Anthropic for the Workspace. Each request is then valued at the tokens it actually used, at the price version frozen on it when it started: uncached input, cache writes (5-minute and 1-hour), cache reads and output are separate counts that do not overlap, and the model’s reasoning counts as output, so a higher effort setting can use more. The price is the provider’s published rate plus a disclosed markup, shown on our pricing page and in Settings → AI usage; speech stays on OpenAI and is valued separately at OpenAI’s price frozen on each request plus the same markup. A month is totalled and rounded once to the cent, with no per-token or per-request minimum. The preview charges $0 and creates no debt; collection is held. Action packs already bought are kept as they are: never converted, deducted or repriced, and existing contracts are not repriced.

Budgets and pauses. In the preview, owners and administrators set a company budget and per-person budgets in US dollars; $0 stops new paid steps. There is no other per-request cap, and use is not unlimited: Ask AI pauses when a budget would be passed, when our own spending limit for the AI service is reached, or when a provider or the model’s context cannot take the request, and it says so. A step that was already authorised finishes and counts. Usage a provider could not confirm keeps its reservation against the budget but is not a confirmed fee. Our own spending limit is ours: it is not an allowance for your Workspace and is never charged to you.

Metered usage. Charging for Ask AI usage needs a separate review by us, the Workspace owner’s acceptance of metered usage, and the 30 days’ email notice before a new fee applies at your next renewal set out in our Terms; it never starts automatically. It is in addition to your Plan, which still provides your modules, users and storage. A confirmed fee for a step never exceeds what was reserved for it; anything above that is absorbed by us.

We may apply reasonable rate limits to keep the Service stable.

11. Switching it off

Workspace owners and administrators can switch the Assistant off for the whole company at any time in Settings. When it is off, nothing is sent to the AI provider. The Assistant is not available to read-only viewers or outside collaborators.

12. Changes to models and providers

AI technology changes quickly. We may change the models or providers behind the Assistant. We will update our Subprocessors page before a new provider begins processing Customer Data, and we will not move to a provider that trains on your content. We will not silently widen what a provider receives or what it is used for: Ask AI uses Anthropic only for a Workspace whose owner has opted in.

13. Disclaimer

The Assistant is provided “as is”. To the fullest extent permitted by law we make no warranty that Output is accurate, complete or fit for any purpose, and the limits of liability in our Terms apply.