AI Use & Data Policy

Document ID: WST-POL-AI-v1.0 · Effective Date: 2 May 2026 · Governing Law: Delaware, USA

This policy explains how artificial-intelligence features in the Workestra platform (the "Service") work, which AI providers Workestra supports, what data is sent where, and how workspace administrators control or disable AI use. It supplements our Privacy Policy and Data Processing Agreement.

1. AI Features in the Service

Workestra's AI capabilities include:

AI features are opt-in. New workspaces start with AI disabled until an administrator configures a provider and API key.

2. Per-Workspace Provider Configuration

Each workspace administrator selects one AI provider for the workspace and provides their own API key, base URL, and model selection. Currently supported providers:

The selected provider is the AI engine for that workspace. If no provider is configured, AI features remain disabled.

3. Data Flow When AI Is Used

When a user invokes an AI feature, Workestra constructs a request and sends it to the configured provider. A typical request includes:

The AI provider returns a response, which Workestra renders in-app and may persist as a conversation log inside the customer's workspace.

4. Workestra Does Not Train on Customer Data

• Workestra does not train, fine-tune, or otherwise improve any AI model using Customer Data.
• We do not pool Customer Data across workspaces for any model-improvement purpose.
• Customer-provided embeddings (used for Cmd+K hybrid search) are stored in pgvector inside the customer's own workspace database and are scoped by workspace_id.

The behavior of the underlying AI provider is governed by your contract with that provider. Most enterprise/API tiers (Anthropic API, OpenAI API, etc.) commit to not using API data for model training; please verify the terms of the provider you select.

5. AI Provider Acts on the Customer's Behalf

When Customer configures the workspace with its own API key for an AI provider, the AI provider acts under the Customer's contract with that provider. The AI provider is therefore not a Workestra sub-processor by default; it is Customer's chosen AI engine. Workestra:

• Transmits data to the provider only as needed to fulfill the user's request
• Does not access, store, or use the API key beyond making authenticated requests
• Lists each supported provider as a conditional sub-processor at /legal/subprocessors for transparency, regardless of contracting model

If you need a different contracting arrangement (for example, a Workestra-procured AI key under a unified contract), contact legal@workestra.app.

6. Write Actions Require Confirmation

Workestra's AI tools are split into read tools (e.g., search, fetch, summarize) and write tools (e.g., create deal, send email, mark ticket resolved). Write tools require explicit user confirmation in the UI before they execute. The MCP server exposes the same tool surface; external clients must run with workspace-scoped tokens, and write actions remain subject to the workspace's RBAC and rate limits.

7. Embeddings and Hybrid Search

To power Cmd+K and similar-records features, Workestra computes vector embeddings of selected workspace content and stores them in the customer's workspace database (pgvector). Embeddings:

• Are scoped by workspace_id and never queried across tenants
• Are deleted when the underlying record is deleted
• Are not shared with third parties except the embedding provider during initial computation

The provider used to compute embeddings is the AI provider configured for the workspace (or a default OpenAI embedding model if not configured otherwise — administrators are notified at configuration time).

8. Output Accuracy and Human Oversight

AI tools can produce inaccurate, outdated, or inappropriate outputs. Customer is responsible for:

• Reviewing AI-generated content before relying on it for business decisions
• Ensuring AI use complies with applicable law in Customer's jurisdictions, including the EU AI Act, US state automated-decision laws, and sector-specific rules
• Configuring access controls so that only authorized users can invoke AI features
• Not using AI features to make solely-automated decisions about natural persons that have legal or similarly significant effects without appropriate safeguards

Workestra does not guarantee specific accuracy levels, response times, or output formats from third-party AI providers.

9. Administrator Controls

Workspace administrators can:

• Disable AI globally for the workspace
• Restrict AI to specific modules or roles (where supported by plan tier)
• Rotate or revoke the AI provider API key
• Audit AI conversation logs through the workspace audit log
• Configure data masking for sensitive fields before they are sent to the provider (where supported)

10. Transparency and Notice

We will document material changes to which providers are supported, what data is sent in standard prompts, or how embeddings are computed. Material policy changes are communicated to administrators by email and in-product notice, with at least 30 days' notice before they take effect.

11. Reporting AI-Related Issues

Policy Changelog