Trust
AI Transparency & Data Use
Kavanah is AI-native. This page explains exactly what data reaches AI providers, how it is (and is not) used, and the controls you have over it.
Last updated: July 1, 2026
AI is central to Kavanah — from the conversational agent to drafting, summarization, and skill-based assignment. Because AI touches customer content, we document its data handling openly rather than burying it in a policy. If your organization has an AI-governance review, this page is written to answer it.
Which providers we use
| Provider | Used for | Data sent |
|---|---|---|
| Anthropic (Claude) | The AI agent, drafting, summarization, and reasoning features | The prompt content for a given request: the messages you send the agent plus the workspace context needed to answer (e.g. the tasks or thread you ask about) |
| OpenAI | Image generation (e.g. AI persona avatars) | The text prompt describing the image to generate |
These providers are also listed on our Sub-processors page.
Your data is not used to train models
Kavanah uses these providers under their enterprise/API terms, which provide that customer content submitted through the API is not used to train their models. Content is processed only to generate the response to your request and is not retained by the provider to improve their models.
Data minimization
The agent operates through a controlled tool-calling loop. It only sends what a request requires — it does not stream your entire workspace to the model. Tools retrieve specific records on demand, and every tool call is subject to the same workspace-scoping and permission checks as the rest of the platform, so the AI cannot reach data the requesting user could not reach directly.
Controls you have
- Capability scopes. Each AI persona can be restricted to specific capability groups, so an agent identity can only touch the surfaces you allow.
- Member access. Personas can be limited to specific members, and administrative approval is required to change who can use a restricted persona.
- Read-only vs. acting. Tools that only read data are classified separately from tools that can make changes.
- Auditability. AI-initiated changes flow through the same immutable audit log as human actions.
Human oversight & accuracy
AI output can be imperfect. Kavanah surfaces AI-generated content as drafts and proposals that a person reviews, and consequential actions remain subject to the acting user's permissions. Customers are responsible for reviewing AI output before relying on it for decisions.
Questions from a security, privacy, or procurement team? Email security@kavanah.ai. For SOC 2, penetration-test, or questionnaire artifacts shared under NDA, use the request forms on the Trust & Compliance page.