AI & Provider Processing

Effective date: July 2, 2026. This page explains how Monocrat uses AI and provider processing while preserving customer-facing billing, privacy, tenant-isolation, source-policy, and high-stakes review commitments.

1. What AI Does In Monocrat

Monocrat uses deterministic software and approved AI systems to ingest, extract, classify, summarize, draft, route, cite, compare, and explain business information. AI work may support customer follow-up, bill and invoice matching, renewal preparation, cash warnings, source-health checks, support replies, review packages, and Trust Graph preparation.

AI output is not itself an external action. High-stakes actions remain subject to source refs, freshness checks, confidence checks, owner review where required, action gates, package or payload hashes, idempotency keys, and audit records.

2. Provider Routing

Monocrat may route approved work across Anthropic, OpenAI, Google Vertex AI/Gemini, cloud model gateways, batch APIs, embedding systems, negotiated providers, or approved open-weight models when the relevant quality, security, latency, privacy, source-policy, tenant-isolation, provider-term, and customer-commitment gates are satisfied.

Candidate or disabled provider routes are planning and evaluation metadata only. A route is not runtime authority until Monocrat has approved the applicable provider agreement, subprocessor posture, data-use and no-training posture, feature-specific retention or logging behavior, deletion or return path, residency commitment, customer disclosure, source-policy fit, and high-stakes action-gate fit.

Customer-facing product surfaces, invoices, Today/Inbox, support, and Billing & Usage show Monocrat usage classes, credits, invoices, receipts, adjustments, and disputes. They do not disclose per-call model routing, actual provider COGS, batch/cache savings, negotiated rates, or internal margin decisions.

3. Customer Data And Model Training

Monocrat does not sell Customer Content. Monocrat does not use Customer Content to train generalized third-party AI models. Provider routes are approved only when provider data-use, retention, deletion, logging, residency, and no-training posture are compatible with Monocrat's customer commitments for the applicable data class.

Monocrat does not enable provider data-sharing, feedback, evaluation, logging, tuning, or dataset-contribution controls that would let a provider use Customer Content for generalized model improvement unless Customer has expressly approved that use in writing. Route approvals must account for provider-specific defaults and feature exceptions, including batch jobs, file APIs, grounding, explicit caching, diagnostics, stateful conversations, abuse monitoring, and zero-data-retention or comparable controls where applicable.

Monocrat may use Customer Content to provide the service to that customer, maintain and secure the service, debug and evaluate quality, generate customer-specific outputs, produce source-backed records, and improve Monocrat systems in ways permitted by the Terms, Privacy Policy, DPA, and applicable law.

4. Google API Data

Google API data is used only for user-facing product functionality, source health, account security, customer-authorized workflows, and legal or security obligations. It is not sold, used for advertising, or used for generalized AI model training. Google Workspace scopes are requested only after approval and through a separate product permission flow.

5. High-Stakes Boundaries

AI may help prepare or explain high-stakes work, but it does not replace professional review or owner approval. Monocrat is not a law firm, accounting firm, tax preparer, payroll bureau, investment adviser, insurance broker, medical provider, money transmitter, bank, or escrow agent. Outputs are not professional advice.

Legal, tax, payroll, payment, refund, public publication, source-system write, account deletion, consent, and customer-facing commitment surfaces may require additional review, manual fallback, or blocking before any external effect.

6. Source References And Missing Data

Monocrat is designed to keep source refs, confidence, currentness, caveats, and dispute state with the work. Missing, stale, conflicting, low-confidence, unavailable, unauthorized, or policy-blocked evidence is treated as a blocker or review gap rather than being silently inferred into a positive claim.

7. Batch And Latency-Tolerant Processing

Monocrat may use batch processing, prompt or context caching, and other latency-tolerant routes for eligible work when selected before the job starts and reflected in product status. Batch or cache savings affect Monocrat's internal provider cost unless a future rate card states otherwise. They do not change the customer-facing usage class, invoice rate, or accepted-work quality requirement.

8. Subprocessors

The AI and cloud providers Monocrat may use are listed at Subprocessors. Provider availability may vary by feature, data class, environment, customer profile, and written order.

Subprocessor listing does not by itself approve every provider feature. For example, a provider may be approved for ordinary prompt/response processing but still require separate review for batch APIs, file storage, grounding, stateful sessions, managed agents, fine-tuning, feedback sharing, or features with different retention, logging, deletion, residency, or human-review behavior.

9. Changes

Monocrat may update this disclosure as provider routes, legal requirements, data-governance requirements, model availability, or customer commitments change.