Structum AI — AI Features and Output Terms
Effective Date: October 1, 2026
Last Updated: October 1, 2026
These AI Terms are part of the Structum AI Terms of Service. They govern every feature of the Services that uses artificial intelligence or automated analysis, and every result those features produce. Capitalized terms not defined here have the meanings given in the Terms of Service (the “Terms”). If these AI Terms conflict with the Terms, these AI Terms control for AI Features and Outputs. The short version: AI Features are tools that help you work faster; they are not a substitute for reading the documents, doing the math, checking the code, or exercising professional judgment. You are responsible for everything you do with an Output.
1. Scope
1.1 What These AI Terms Cover
These AI Terms apply to all AI Features, meaning any functionality of the Services that uses machine learning, large language models, generative AI, optical character recognition, natural-language processing, computer vision, statistical or rules-based inference, retrieval-augmented generation, or other automated techniques to extract, classify, search, summarize, answer, match, compare, calculate, draft, recommend, prioritize, or take or propose actions. They also apply to all Outputs, meaning any result an AI Feature produces, including extracted fields and line items, classifications, summaries, chat responses and citations, quantities and takeoffs, estimates and pricing, bid comparisons and leveling sheets, vendor matches and compliance indications, schedules and critical-path calculations, code and permit information, extracted contract terms, financial classifications and analyses, drafted communications and documents, recommendations, and proposed or prepared actions.
1.2 AI Clients
These AI Terms also apply to Customer’s use of third-party AI Clients connected to the Services through the MCP Server or API, to the extent stated in Section 7.
1.3 Who Is Bound
Customer is bound by these AI Terms and is responsible for ensuring that every Authorized User who uses an AI Feature, or who acts on an Output, understands and complies with them.
2. How AI Features Work, and What That Means
2.1 Probabilistic, Not Deterministic
AI Features generate Outputs by applying statistical models to Customer Content and to the user’s prompts and settings. The same input can produce different Outputs at different times, and an Output that is fluent, confident, and well-formatted can nonetheless be wrong. AI Features can misread, mislocate, or miss information that is in Customer Content; can state as fact things that are not in Customer Content or are not true; can produce numbers, dates, names, and citations that do not exist; and can be internally inconsistent. These behaviors are inherent in the technology and are not defects that Structum can guarantee to eliminate.
2.2 Grounding and Citations
Where an AI Feature cites a source page or record, the citation identifies where the model retrieved information from; it does not guarantee that the Output accurately reflects that source, that the source is the current or governing version of the document, or that other relevant sources were not missed. Customer must open and read the cited source before relying on the Output.
2.3 Inputs Determine Outputs
AI Features work only from the documents, data, and instructions available to them in Customer’s environment and from the permissions Customer has configured. Outputs may be wrong or incomplete if Customer Content is missing, illegible, mislabeled, out of date, superseded by an addendum or revision, duplicated, or stored in a project or folder the AI Feature was not given access to. Customer is responsible for maintaining Customer Content so that current documents are identifiable as current.
2.4 No Review by Structum
Structum does not review, verify, approve, or edit Outputs before they are presented. An Output is not a statement, representation, or advice by Structum. Structum has no obligation to monitor Customer’s use of Outputs.
2.5 Models Change
Structum may change the models, providers, prompts, retrieval methods, and settings behind any AI Feature at any time, including in response to provider deprecations, and Outputs may change as a result. Structum does not commit to any particular model, provider, version, or behavior.
2.6 Units, Formats, and Conventions
AI Features may misinterpret units of measure, scales, currencies, date formats, cost codes, CSI divisions, drawing conventions, abbreviations, and industry shorthand, and may apply a convention different from the one Customer intended. Customer must confirm units, scales, and conventions in every quantitative Output.
3. Customer’s Duty to Review and Verify
3.1 General Standard
Before relying on, acting on, distributing, or incorporating any Output into any bid, proposal, estimate, budget, contract, purchase order, change order, pay application, invoice, lien waiver, schedule, submittal, RFI, transmittal, meeting minutes, compliance determination, filing, communication, or decision, Customer will review the Output with the same care it would apply to work prepared by an unlicensed junior employee working from the same documents: check it against the source documents, verify the arithmetic, confirm the applicable requirements, and have a qualified person approve it.
3.2 Licensed Professional Review
Where the subject matter of an Output is one that law, contract, professional standards, or ordinary prudence requires a licensed or qualified professional to prepare or review—including structural, mechanical, electrical, plumbing, fire-protection, life-safety, accessibility, and civil matters; code compliance; permit applications; engineered means and methods; legal rights and obligations; lien and bond matters; tax, accounting, and financial reporting; and insurance—Customer will obtain that review before relying on the Output. Outputs are not sealed, stamped, certified, or signed by any professional and must not be presented as if they were.
3.3 Financial and Commitment Controls
Customer will maintain its own approval controls for any Output that affects money or commitments, including estimates, bids, purchase orders, change orders, pay applications, invoices, and lien waivers, and will not permit an Output to commit Customer or release a payment or waiver without approval by a person authorized to do so under Customer’s own procedures.
3.4 Records of Review
Customer is responsible for keeping whatever record of its review and approval of Outputs its own policies, contracts, or laws require. The Services’ audit logs are provided for Customer’s convenience and are not a substitute for Customer’s own approval records.
3.5 Reporting Errors
Customer is encouraged to report material Output errors to support@structumai.io, with the source document and the incorrect Output, so that Structum can investigate. Reporting does not create any obligation of Structum to correct the Output or any liability for it.
4. Feature-Specific Terms
The following table describes the principal AI Features, their known limitations, and what Customer must do before relying on their Outputs. It is illustrative, not exhaustive; new features are subject to these AI Terms whether or not listed. The absence of a limitation from this table does not mean the limitation does not exist.
| AI Feature | What it does | Known limitations | Customer must |
|---|---|---|---|
| Document extraction and classification (proposals, bids, invoices, receipts, certificates of insurance, licenses, permits, vendor documents, emails) | Identifies document type and extracts fields, line items, amounts, dates, parties, and terms into structured records. | Misreads handwriting, low-resolution scans, stamps, rotated or multi-column layouts, and tables; confuses similar fields (subtotal vs. total, bid vs. alternate, expiration vs. effective date); merges or drops line items; misclassifies documents; cannot detect forged or altered documents; does not verify that a certificate or license is genuine, current, or adequate. | Compare every extracted value to the source document before it is used in any record, payment, or compliance determination; independently verify insurance, license, and certification status with the issuer where it matters. |
| Project chatbot and document Q&A | Answers questions using retrieval from Customer’s project documents, with source-page citations. | May answer from the wrong document version, from a superseded drawing or specification, or from a document that is not the governing contract document; may omit relevant documents; may generalize beyond the cited source; may not know what the documents do not say; answers to prior questions may be reused and may be stale. | Open and read the cited pages; confirm the cited document is current and governing; treat “I could not find” answers as a prompt to search manually, not as confirmation that something does not exist. |
| Summarization, meeting minutes, action items, weekly reports | Summarizes documents, transcripts, communications, and project activity; extracts action items and decisions. | May omit material points, misattribute statements to the wrong speaker or party, misstate decisions or commitments, invent action items, or mis-assign owners and dates; transcripts may contain transcription errors that propagate into minutes. | Review minutes against the recording or notes before distribution; confirm any commitment, direction, or notice attributed to an owner, architect, or subcontractor before acting on it or relying on it as a contractual communication. |
| Plan, specification, and addendum analysis | Reads drawings and specifications to identify scope, notes, schedules, and changes; compares addenda and revisions. | Does not reliably interpret graphical content, symbols, details, sections, dimensions, revision clouds, or cross-references; may miss scope on sheets it does not analyze; may not detect conflicts between drawings and specifications or between disciplines; addendum comparison may miss changes. | Verify scope, dimensions, and revisions on the drawings themselves; perform its own document coordination and conflict review; confirm that every addendum has been received and incorporated. |
| Takeoffs and quantities | Estimates quantities of materials, areas, lengths, counts, and assemblies from drawings and specifications. | Depends on correct scale, legibility, and drawing completeness; may double-count, omit, or misclassify items; does not account for waste, lap, overage, phasing, site conditions, or means and methods; unit conversions may be wrong. | Independently verify quantities and scale; apply its own waste, overage, and productivity factors; treat AI quantities as a starting point, never as the basis for a committed price. |
| Estimating, cost breakdowns, and pricing | Builds estimates and cost breakdowns from quantities, templates, historical data, and vendor pricing; validates estimate structure; compares estimate versions. | Unit costs and productivity assumptions may be outdated, regional, or inapplicable; markups, taxes, bonds, insurance, general conditions, escalation, and contingencies may be missing or wrong; version comparisons may misalign line items; validation checks are structural, not a review of pricing judgment. | Price every bid and estimate on its own judgment; verify current market pricing; review all markups and exclusions; confirm arithmetic and totals before issuing any bid, proposal, guaranteed maximum price, or budget. |
| Bid leveling and comparison | Normalizes vendor bids to a common format, identifies scope gaps, inclusions, exclusions, and alternates, and compares pricing. | Scope gap detection depends on how the AI interprets bid language and may miss exclusions, qualifications, or conditions; may misalign line items across bidders; may mis-total; does not assess bidder qualifications, capacity, bonding, or risk. | Read each bid in full; confirm scope coverage and exclusions directly with bidders; make award decisions on its own evaluation; comply with all procurement rules applicable to the project. |
| Vendor matching, vendor setup, and vendor compliance | Suggests vendors for a project or scope from Customer’s vendor data and other sources; extracts vendor facts; tracks insurance, license, and document expirations. | Matches are based on available data and pattern similarity, not on verified qualifications, licensure, insurance, safety record, financial condition, or performance; extracted vendor facts may be wrong; expiration tracking depends on correct extraction and on documents Customer has uploaded; compliance indications are not a determination that a vendor meets any contractual or legal requirement. | Perform its own prequalification and due diligence; verify licenses and insurance with the issuing authority or carrier; make its own decisions about whom to invite, award, and pay; not treat a compliance indication as a substitute for reviewing the underlying certificate against the contract requirements. |
| Scheduling and critical-path calculations | Generates schedules, activities, durations, sequences, and critical-path and float calculations; suggests calendar events and follow-ups. | Durations and logic are generic assumptions; may omit activities, constraints, lead times, weather, permitting, inspections, or resource limits; critical-path and float results are only as good as the logic; generated dates are not commitments and are not a baseline or an as-built schedule. | Build and maintain its own project schedule with a qualified scheduler; verify logic and durations; issue no schedule, notice of delay, or time-related claim based solely on an Output. |
| Construction-code lookup, permit requirements, and compliance analysis | Retrieves and summarizes code provisions, permit requirements, and compliance considerations; tracks permit status and comments. | Sources may be outdated, incomplete, or the wrong edition; local amendments, jurisdictional interpretations, and authority-having-jurisdiction requirements vary and may not be reflected; summaries may misstate provisions; compliance analysis is not a code review, inspection, or approval and does not consider all applicable requirements. | Confirm all code, permit, and inspection requirements with the authority having jurisdiction and with licensed design professionals; not represent to any owner, authority, or third party that an Output constitutes a code determination. |
| Contract term extraction and contract management | Extracts contract sums, dates, retainage, payment terms, notice provisions, scopes, and exclusions; tracks change orders and pay applications against contracts. | May misread or omit terms; may not capture amendments, exhibits, incorporated documents, or handwritten changes; may misstate notice deadlines and conditions precedent; calculated balances may be wrong if inputs are wrong. | Rely on the executed contract documents, not extracted terms, for any notice, claim, payment, or dispute; verify all balances against its own accounting records. |
| Pay applications, invoices, purchase-order import, and accounting classification | Prepares pay applications and schedules of values; extracts invoice data; classifies line items and assigns cost codes or titles; prepares records for accounting integrations. | Classifications and cost codes are suggestions and may be wrong; amounts, retainage, and previously billed values may be miscalculated; the Services do not verify that work was performed, that amounts are owed, or that a payment is due; imported records may duplicate or conflict with records in Customer’s accounting system. | Review every classification and amount with accounting personnel; verify work-in-place and entitlement before certifying or paying; reconcile all synchronized records in the destination system; retain sole responsibility for all payments and financial statements. |
| Lien waivers, notices, and statutory forms | Generates or pre-fills lien-waiver, notice, and similar forms from project data. | Statutory form requirements, notarization, conditional/unconditional distinctions, timing, and delivery rules vary by state and by project type and may not be reflected; pre-filled amounts and dates may be wrong; the Services do not determine whether a waiver or notice is effective or enforceable. | Have counsel or a qualified person confirm the correct form, content, timing, and execution requirements for the project’s jurisdiction before issuing or accepting any waiver or notice. |
| Drafted communications and automated follow-ups (emails, texts, RFIs, submittal transmittals, bid follow-ups) | Drafts communications from project context; sends follow-ups to bidding vendors and others on a schedule Customer configures. | Drafts may misstate facts, dates, amounts, or requirements, may use an inappropriate tone, may omit or add recipients, and may be sent to the wrong party if Customer’s contact data is wrong; automated follow-ups continue on their schedule until Customer stops them; a drafted communication may have contractual or legal effect once sent. | Read and edit every draft before sending; confirm recipients; treat every sent message as Customer’s own; monitor and stop automated follow-ups when they are no longer appropriate. |
| Recommendations, prioritization, and risk flags (overdue and missing submittals, pending items, suggested next steps, project search results) | Highlights items that appear overdue, missing, inconsistent, or important, and suggests actions. | Flags depend on data completeness and configured dates; missing flags do not mean nothing is missing; suggested actions may be wrong for the situation; search results may omit relevant records. | Maintain its own logs, registers, and tracking; treat flags as prompts for review, not as a complete or authoritative list; never rely on the absence of a flag. |
5. Assisted and Automated Actions
5.1 Previews and Confirmation
Some AI Features, and AI Clients connected to the Services, can prepare actions such as creating a purchase order from a proposal, creating a change order or RFI, generating a schedule, importing purchase-order line items, sending vendor follow-ups, or updating records. Actions that would send a communication outside Customer’s organization, commit or move money, or make a change that cannot be undone are presented as a preview and take effect only when an Authorized User confirms them in the Services. Structum may add, change, or remove confirmation requirements, and may require confirmation for additional categories of action, at any time.
5.2 Confirmed and Configured Actions Are Customer’s
An action confirmed by an Authorized User is Customer’s action, exactly as if the Authorized User had performed it manually. If the Services permit Customer to configure an automation to run without individual confirmation (for example, scheduled bid follow-ups), every action taken under that configuration is Customer’s action from the moment Customer enables it until Customer disables it. Customer is responsible for reviewing the preview, for the accuracy of the underlying data, for the consequences of the action, and for monitoring and stopping automations.
5.3 Reversal Not Guaranteed
Structum does not guarantee that any action can be reversed, recalled, or corrected after it takes effect, including a sent communication, a record synchronized to a Third-Party Service, or a change visible to collaborators.
5.4 Failed or Partial Actions
An action may fail, time out, or complete only in part because of a provider outage, integration error, permission change, or other cause. Customer will verify that each action it intended to take was completed, and will not assume that an action was taken because it was requested.
6. Data Used by AI Features, Model Providers, and Training
6.1 What AI Features Use
AI Features process the prompts, selected documents, Customer Content, connected-account data, retrieved project context, prior interactions, settings, and technical metadata reasonably necessary to produce the requested Output, subject to the permissions Customer has configured. Structum seeks to transmit only what is reasonably necessary for the feature.
6.2 Model Providers
Structum uses third-party model, hosting, transcription, document-processing, and AI-infrastructure providers (“Model Providers”) to deliver AI Features, and may route requests among them based on availability, capability, cost, security, and configuration. Model Providers are Structum’s Subprocessors and are engaged under terms that, where available, prohibit them from using Structum’s inputs and outputs to train their generalized models and limit their retention. Structum will identify its material Model Providers in its Subprocessor List or on request to privacy@structumai.io. Structum is not responsible for a Model Provider’s outage, degradation, content filtering, refusal to process a request, model change, or deprecation.
6.3 No Training on Customer Content Without Permission
Structum does not use Customer Content or Outputs to train or fine-tune generalized AI models made available to other customers or the public, unless Customer expressly opts in or an Order Form provides otherwise. Structum may use Customer Content within Customer’s environment to create indexes, embeddings, retrieval stores, and Customer-specific configurations that improve results for Customer, and may use Usage Data, de-identified data, and feedback signals (such as ratings, edits, and corrections) to improve AI Features in accordance with the Terms and the Privacy Notice.
6.4 Human Access
Authorized Structum personnel and contractors may access AI inputs and Outputs only when reasonably necessary to provide support Customer requests, investigate a security, abuse, or defect report, comply with law, or conduct a quality review permitted by the Agreement, subject to confidentiality obligations and access controls.
6.5 Prompt Injection and Untrusted Content
Documents, emails, and other content processed by AI Features may contain text that attempts to instruct an AI model (for example, a vendor proposal containing hidden instructions). Structum implements safeguards intended to reduce this risk but cannot eliminate it. Customer will treat Outputs derived from third-party documents with corresponding caution and will not configure automations that act on third-party content without review.
7. AI Clients and the MCP Server
7.1 Customer’s Choice, Customer’s Responsibility
Customer may connect an AI Client of its choosing to the Services. The AI Client’s provider is an independent third party selected by Customer, not Structum’s Subprocessor, and Customer’s use of the AI Client is governed by the provider’s terms. Structum does not control the AI Client’s reasoning, statements, or proposed actions, its provider’s data retention or model-training practices, or its security.
7.2 Permissions and Least Privilege
The MCP Server enforces the permissions Customer grants to a connection on a per-module basis. Customer will grant an AI Client only the modules and projects it needs, will review connections periodically, and will revoke connections that are no longer needed. Customer is responsible for all data an AI Client retrieves within the permissions Customer granted.
7.3 Outputs Produced by AI Clients
Content that an AI Client generates from data retrieved through the Services is the AI Client’s output, not Structum’s Output, and Structum makes no representation about it. Sections 2, 3, 5, and 9 nonetheless apply to Customer’s reliance on any such content to the same extent as to Outputs.
7.4 Actions Through AI Clients
Section 5 applies to every action an AI Client proposes or initiates. Structum’s confirmation controls operate within the Services; they do not control what the AI Client does outside the Services with the data it retrieved.
7.5 Directory Listings
If Structum’s connector is listed in a third-party AI directory or marketplace, the listing is a convenience. The directory operator and the AI Client provider are not Structum’s agents or Subprocessors, and their terms do not bind Structum.
8. Ownership of Outputs
8.1 Customer Owns Outputs
As between Customer and Structum, Customer owns the Outputs generated for Customer, and Structum assigns to Customer any right Structum may have in them, subject to Section 8.2 and to Structum IP’s ownership of the Structum Platform, models, prompts, templates, workflows, and methods used to generate Outputs, which are not assigned or licensed.
8.2 No Exclusivity or Uniqueness
AI Features generate Outputs from patterns learned from large data sets and from templates common to all customers. Outputs generated for other customers may be similar or identical to Customer’s Outputs. Structum does not represent that any Output is unique, original, protectable by copyright or other intellectual-property law, or free of third-party claims, and Customer’s ownership does not extend to any element of an Output that is common to Outputs generated for others.
8.3 Third-Party Material in Outputs
Outputs may reproduce, summarize, or derive from material in Customer Content that belongs to third parties, such as drawings prepared by design professionals, vendor proposals, and specifications. Customer’s rights in such Outputs are limited by the rights of those third parties, and Customer is responsible for having the rights it needs to use and distribute them.
9. Prohibited Uses of AI Features
Customer will not use AI Features or Outputs, and will not connect an AI Client to the Services, to:
make, or serve as the sole or principal basis for, any decision that produces legal or similarly significant effects concerning an individual, including decisions about employment, hiring, termination, discipline, compensation, promotion, work assignment, housing, credit, lending, insurance, education, healthcare, or access to essential goods or services;
evaluate, score, rank, profile, or draw inferences about any individual’s protected characteristics, union activity, immigration status, health, or eligibility for any benefit;
determine, without review by a licensed professional, any structural, life-safety, fire-protection, accessibility, hazardous-materials, or other matter on which the safety of persons or property depends;
prepare any filing, certification, sworn statement, lien waiver, notice, or representation to a court, government authority, owner, surety, or lender without human review and approval;
generate content that is deceptive, defamatory, harassing, or that impersonates any person or organization, or that misrepresents an Output as the work or certification of a licensed professional;
facilitate bid rigging, price fixing, collusion, or any other anticompetitive conduct, including by sharing competitors’ pricing obtained through the Services;
circumvent, or attempt to circumvent, any Model Provider’s usage policies, the Services’ safety controls, or the confirmation requirements in Section 5;
extract, reconstruct, or reverse engineer any model, prompt, weight, or system component, or use Outputs to train or improve any AI model other than as permitted by the Terms; or
process any data category prohibited by the AUP.
10. Regulatory Responsibilities
10.1 Customer Is the Deployer
Customer, not Structum, decides how AI Features and Outputs are used in Customer’s business and projects. To the extent any law regulating the use of artificial intelligence or automated decision-making technology—including state laws governing consequential decisions, the California Consumer Privacy Act regulations on automated decision-making technology, and, where applicable, the EU Artificial Intelligence Act—imposes obligations on the person that deploys or uses an AI system, Customer is that person with respect to Customer’s use of the Services, and is responsible for any required notice, impact or risk assessment, opportunity to opt out or correct, human review, appeal, record-keeping, or disclosure to individuals.
10.2 Intended Use
The Services are general-purpose construction project-management tools and are not designed, intended, or marketed for use in making consequential decisions about individuals. Structum designs AI Features to assist Customer personnel, not to replace their judgment, and expects Customer to use them in that way.
10.3 Documentation
On Customer’s reasonable written request, Structum will provide the information about an AI Feature’s intended use, general capabilities and limitations, data categories used, and appropriate oversight that Structum makes generally available to customers, to assist Customer in meeting its own legal obligations. Structum may provide this information through an AI transparency notice, the Documentation, or the Services.
10.4 Professional and Industry Rules
Customer is responsible for compliance with any professional-licensing, procurement, public-works, prevailing-wage, safety, or industry rule that limits or conditions the use of automated tools in its work.
11. Human Oversight, Feedback, and Corrections
11.1 Human in the Loop
Customer will ensure that a qualified person reviews Outputs as described in Section 3, that Authorized Users understand the limitations of the AI Features they use, and that Customer’s workflows do not permit an Output to become a commitment, communication, or record of Customer without the review its own procedures require.
11.2 Ratings, Edits, and Corrections
Customer may rate, edit, correct, or comment on Outputs in the Services. Corrections update Customer’s records within its environment; they do not retrain the underlying models. Structum may use ratings, edits, and corrections as feedback to improve AI Features in accordance with Section 6.3.
12. Availability, Changes, and Deprecation
Structum may add, modify, suspend, limit, or discontinue any AI Feature, or change its underlying Model Provider, model, prompts, or settings, at any time, with or without notice, including where required by a Model Provider’s deprecation or policy. AI Features may be unavailable, rate-limited, or degraded from time to time. Structum may impose usage limits on AI Features and may charge for usage above included allowances as described in the applicable plan or Order Form.
13. Disclaimers and Limitation of Liability
Sections 17 (Disclaimers), 18 (Indemnification), and 19 (Limitation of Liability) of the Terms apply to AI Features and Outputs and are incorporated here. Without limiting them: AI FEATURES AND OUTPUTS ARE PROVIDED “AS IS.” STRUCTUM AND ITS AFFILIATES, LICENSORS, AND MODEL PROVIDERS DO NOT WARRANT THE ACCURACY, COMPLETENESS, CURRENCY, RELIABILITY, OR SUITABILITY OF ANY OUTPUT; DO NOT WARRANT THAT ANY AI FEATURE WILL IDENTIFY EVERY ERROR, OMISSION, CONFLICT, DEADLINE, REQUIREMENT, OR RISK; AND ARE NOT LIABLE FOR ANY LOSS ARISING FROM RELIANCE ON AN OUTPUT OR FROM ANY ACTION CONFIRMED, CONFIGURED, OR INITIATED BY CUSTOMER, AN AUTHORIZED USER, OR AN AI CLIENT. CUSTOMER ASSUMES ALL RISK OF USING OUTPUTS.
14. Contact
Questions about these AI Terms: legal@structumai.io. Reports of material Output errors: support@structumai.io. Security concerns, including suspected prompt-injection or abuse: security@structumai.io.