Legal
Privacy Policy
Praxeum ("Praxeum", "we", "us") is a service operated by AgentLabs, registered with the Netherlands Chamber of Commerce (KvK) under number 42041148, the data controller for this website and the Praxeum application. This policy explains what data Praxeum accesses, how we use and protect it, and the rights you have. Questions or requests: bram@agentlabs.nl.
1. What Praxeum is
Praxeum helps a company turn its own knowledge into verified, human-approved "skills" that its AI agents can execute safely. To do this, a customer connects sources they control (such as Slack, Google Workspace, and Notion) and explicitly scopes which channels, drives, folders, and meeting series Praxeum may read. Praxeum processes this content on the customer's behalf. In relation to a customer's own workspace data, Praxeum acts as a data processor and the customer is the controller; AgentLabs is the controller for account, billing, and website data.
2. Google user data we access
Where your organization connects Google Workspace, Praxeum requests one read-only Google scope:
| Scope | Why we request it |
|---|---|
https://www.googleapis.com/auth/drive.readonly | To read the Google Docs in the shared drives and folders your organization explicitly selects, together with each file's title, folder location, and sharing permissions, so we can convert procedure-shaped documents (SOPs, runbooks, checklists) into cited draft skills for your team, and show each person only the documents they can already open in Google Drive. |
What we access. The text content of Google Docs in the selected shared drives and folders, including their unresolved comments (comment text and author); file metadata (title, file type, folder location, last-modified time, link); and each file's sharing permissions (the email addresses and roles of the people it is shared with). For other file types in those locations we store only the title, type, link, and permissions, not the content. We do not access Gmail, Calendar, Contacts, or any Drive location your organization has not selected.
How we use it. Only to provide the user-facing Praxeum features described on this page: drafting cited skills that a person in your organization reviews and approves, answering your team's questions with citations, and enforcing the same access your Google Drive sharing already grants. To draft skills and build search, document text is processed by large language models and embedding models running on Amazon Bedrock in an EU region (see Subprocessors).
How we share it. We do not sell Google user data and do not share it with anyone except the subprocessors listed below, and only as needed to run the service: Nango (stores the Google access token), Amazon Web Services (model processing), Supabase and Railway (storage and hosting). Error and observability tools receive scrubbed or redacted data only.
We do not request write access to Google data. Praxeum never fetches content outside the drives and folders you select. If we add Google Meet transcript import, we will request the separate drive.meet.readonly scope and update this policy first.
3. How we use data
- To mine, draft, and serve verified skills within your organization only.
- Every skill is drafted with citations to its source and requires human review and approval before use.
- We never produce individual employee performance analytics; source scoping is visible to employees by design.
- Content you connect is not used to train or fine-tune any AI or ML model. Reviewer edits are stored within your own tenant only.
4. How we store and isolate data
- Data is logically isolated per customer tenant, enforced in the database (row-level security) as well as in application code.
- Personal data and secrets (e.g. API keys, national identifiers) are redacted at ingestion before content is stored or served.
- Data is encrypted in transit and at rest by our infrastructure providers.
- Google OAuth tokens are held by our connection provider (Nango), never in our own database or logs.
5. Subprocessors
We use a short, disclosed set of subprocessors to operate the service:
| Subprocessor | Purpose |
|---|---|
| Nango | Source connection auth & sync |
| WorkOS | Authentication & agent authorization |
| Amazon Web Services (Amazon Bedrock, EU region) | LLM processing with Anthropic Claude models (mining & serving) and text embeddings with Cohere models |
| LlamaParse (LlamaIndex) | Document-to-text parsing for PDF and Office files (planned; not yet in use) |
| Langfuse | LLM observability (redacted payloads) |
| Sentry | Error monitoring (payloads scrubbed) |
| Resend | Transactional email |
| Railway / Supabase | Application hosting & database |
6. Retention & deletion
- When a source item is deleted or a source is de-scoped, we propagate the removal to derived documents, chunks, and embeddings. For Google Drive, a file that is deleted, trashed, or moved out of the selected drives and folders is removed from Praxeum, normally within 24 hours.
- You can revoke Praxeum's access to Google data at any time from your Google Account (myaccount.google.com/permissions) or, for your organization, from the Google Admin console. Revoking stops all further reads. To have data that was already read deleted, contact us; we delete it with a full tenant purge.
- On account offboarding we run a full tenant purge across all data and revoke source connections.
- You may request erasure of personal data; approved-skill evidence retained for audit is subject to an erasure override on request.
7. Your rights
Depending on your location (including under the EU/UK GDPR), you may have rights to access, correct, delete, restrict, or port your personal data, and to object to processing. To exercise these, contact bram@agentlabs.nl. For workspace data processed on a customer's behalf, requests are directed to that customer as controller and we assist as processor.
8. International transfers
We aim to process data in the EU where possible and use providers offering EU data regions. Where data is transferred outside the EEA, appropriate safeguards (such as Standard Contractual Clauses) apply.
9. Changes
We will update this policy as the service evolves and revise the "last updated" date above.