Legal

Claude AI usage & compliance.

How RevSure uses Anthropic Claude through Google Cloud Agent Platform, and the controls that govern that usage.

Overview

RevSure Lite is an AI assistant that enables authorized users to explore their RevSure marketing and revenue data through natural conversation and interactive visual dashboards. Its conversational AI, Reli, is powered by Anthropic Claude running on Google Cloud Agent Platform (formerly Vertex AI) via the Claude Agent SDK.

Depending on the user's question, RevSure Lite may call multiple RevSure data tools in sequence to retrieve and analyze the relevant authorized data. The AI reasons over the retrieved results and determines the appropriate response format: a direct conversational answer, an interactive artifact such as a rendered dashboard, chart, or data visualization, or both, allowing the user to explore the results interactively.

The service is designed to operate within each customer's existing authorization boundaries, contractual restrictions, and RevSure's security and privacy requirements.

How RevSure Lite works

  • The user authenticates and submits a question or request through the RevSure Lite interface.
  • RevSure validates the user's identity, tenant membership, and permitted data scope. Tenant identity and authorization context are passed with every request — the AI layer never determines its own access scope.
  • RevSure Lite invokes the Claude Agent SDK, which runs a conversational agent ("Reli") with access to RevSure's MCP (Model Context Protocol) tool set — a curated set of data-retrieval and analysis tools scoped to the authenticated tenant's authorized data.
  • The agent calls the relevant RevSure data tools as needed (e.g. funnel metrics, campaign attribution, pipeline data), gathering only the data required to answer the question.
  • A classification step determines whether the response warrants a plain conversational answer or an interactive visual artifact (dashboard, chart, data table).
  • If an artifact is warranted, a separate, dedicated code-generation step produces a self-contained interactive React component rendered within a controlled sandbox. The sandbox has no access to RevSure's systems or external networks at runtime.
  • The final response — conversational answer, artifact, or both — is returned to the authenticated user.

Note: For complex queries requiring large data payloads, RevSure Lite may temporarily write tool-result data to disk within the session's own isolated working directory on the processing node. This data is scoped to the individual session and is not shared across tenants or retained beyond the session's lifetime, subject to the retention schedule described below.

Legal and customer-usage principles

  • Data protection. RevSure processes customer data in accordance with applicable agreements, its privacy and security program, and applicable data-protection laws.
  • Purpose limitation. Data is used only as needed to provide, secure, and support the requested service and to meet applicable obligations. RevSure Lite data is not used for advertising or unrelated purposes.
  • Data minimization. RevSure Lite is designed to retrieve and send Claude only the data required for the specific query. The MCP tool layer enforces tenant-scoped access; the AI agent cannot retrieve data outside the authenticated user's authorized scope.
  • Transparency. RevSure discloses the use of AI and maintains governance over data use, model changes, and output quality.
  • Security. RevSure applies access control, encryption, logging, monitoring, and incident-management controls appropriate to the service and data sensitivity.

Model and infrastructure

RevSure Lite accesses the Claude Agent SDK, routed through Google Cloud Agent Platform (formerly Vertex AI) using the CLAUDE_CODE_USE_VERTEX=1 configuration. Authentication uses Google Cloud Application Default Credentials; RevSure's own API key is not passed directly to Anthropic.

Model inference occurs within RevSure's Google Cloud project. RevSure does not operate or maintain the Claude model itself; Anthropic maintains the model and Google Cloud maintains the inference infrastructure.

claude-sonnet-4-6 is the default model selection. Users may select an alternative Claude model and adjust the effort level (which controls reasoning depth and response time) based on their preference and the complexity of their query.

Model training and data use

RevSure does not use customer-provided data to train Claude. Google Cloud's published terms state that it does not use customer data to train or fine-tune managed AI/ML models without the customer's prior permission or instruction. Anthropic's commercial terms for Claude on Agent Platform state that Anthropic may not train models on Customer Content from the service.

RevSure uses Claude only to perform authorized AI-assisted functions within RevSure Lite and to operate, secure, and support the service.

Data that may be sent to Claude

The data included in a request depends on the user's question, their authorization scope, and which RevSure data tools are invoked. It may include:

  • The user's prompt and limited prior conversation context required for the current request.
  • Tenant configuration and calendar context (e.g. fiscal calendar, funnel stage definitions, ICP definitions) used to ground the model's reasoning.
  • Aggregated KPIs, metrics, and calculations derived from authorized RevSure data.
  • Campaign, channel, region, lead, account, pipeline, and revenue metadata retrieved in response to the query.
  • High-level behavioral or buyer-journey information relevant to the requested analysis.
  • Limited professional identity data — such as name, work email, company, or account owner — when necessary and within the user's authorized data scope.
  • For skill-guided analyses requiring complete datasets, raw tool-result data may be written to a session-scoped temporary file and processed locally rather than passed in full to the model.

Data exclusions and minimization

RevSure Lite does not intentionally send passwords, API keys, access tokens, payment-card data, government identifiers, health information, special-category personal data, or data outside the user's authorized tenant and role. Users must not enter such information unless RevSure has expressly approved the use case.

The MCP tool layer enforces tenant isolation: each request is scoped to the authenticated tenant's data only. RevSure applies Bash execution sandboxing and network isolation controls within the AI processing layer to prevent in-session data exfiltration to unauthorized external destinations.

Where supported, RevSure may apply field restrictions, masking, or redaction before data is included in model context. Excluding information may reduce the relevance or accuracy of the response.

Security and access controls

  • Access to RevSure Lite is authenticated and governed by the user's tenant membership and assigned permissions, supplied by the RevSure Java backend on every request. The AI layer does not independently manage identity or access.
  • RevSure uses Google Cloud identity and access controls to restrict which workloads can invoke the Claude model endpoint.
  • Google Cloud encrypts customer content at rest by default and in transit.
  • RevSure applies least-privilege access, tenant isolation, monitoring, incident response, and controlled rollout for AI model and configuration changes.
  • The artifact rendering sandbox is network-isolated at runtime — generated dashboards cannot make outbound calls to RevSure systems or external services.
  • RevSure monitors for unauthorized access, excessive data retrieval, prompt injection, unsafe model output, and suspected data exposure.

Retention and logging

RevSure manages retention of RevSure Lite conversations, session data, operational logs, and supporting records in accordance with approved retention schedules and customer agreements. Retention periods may depend on enabled product features and Agent Platform configuration.

Session-scoped temporary files (used for large tool-result processing) are retained only for the duration of the active session and are not persisted to shared storage.

A zero-retention or no-logging statement should be published only after RevSure has verified the behavior of the production API, request/response logging, conversation storage, application logs, caches, and backup lifecycle in the current deployment configuration.

Access to retained information is restricted to authorized RevSure personnel.

Responsible AI and user control

  • AI-generated responses and artifacts may be inaccurate, incomplete, or based on imperfect data and should be verified before material use.
  • Human review is required before an output is used for a material legal, financial, employment, security, customer-facing, or contractual decision.
  • RevSure tests RevSure Lite for accuracy, tenant isolation, prompt injection resistance, harmful content, and unauthorized data exposure.
  • Where supported, customers may work with RevSure to restrict selected AI features, data fields, or artifact generation capabilities.
  • Users and customers have a channel to report inaccurate, unsafe, privacy-impacting, or security-related responses.
  • Users may select from available Claude model versions and adjust the effort level for their session. Higher effort levels apply more reasoning to complex queries at the cost of increased response time; the default configuration is suitable for most use cases.

Customer control and governance

RevSure retains responsibility for approving RevSure Lite use cases, managing user access, reviewing model and configuration changes, evaluating output quality, responding to incidents, and keeping customer-facing disclosures current.

RevSure should review this document at least annually and whenever the model version, endpoint configuration, data scope, retention behavior, customer controls, contractual terms, or applicable law materially changes.

References