Data Privacy & Security

LlamaParse Platform provides data-protection controls and contractual agreements to support enterprise document processing.

Model training and data sharing

By default, we do not use customer data for model training.

If an organization administrator explicitly opts in to our data-sharing program, we may retain and use that organization’s LlamaParse data for model training under the program’s terms. Participation is optional.

Retention requirements

Files uploaded for Parse or Extract are cached for 48 hours to avoid duplicate processing costs, then deleted. You can:

• Delete data on demand at any time through the API or platform UI.
• Set do_not_cache=true on a request to skip caching entirely.

Data processed into an index — parsed content, embeddings, and metadata — is retained separately, for the duration of your agreement to support retrieval. It is not subject to the 48-hour cache window above, is never used for training, and is deleted only when you request deletion or upon contract termination or expiration.

For deployment-specific retention or deletion questions, contact us.

Managed service encryption

• At rest: customer files are encrypted using AES-256.
• In transit: external connections use TLS 1.2 or higher.

Customer-managed encryption keys (BYOK) are not currently supported for our managed service.

LLM providers and subprocessors

We maintain zero-data-retention and no-training agreements with the third-party LLM providers used by our managed service, including US deployments. These agreements cover the providers’ handling of data during processing.

For our subprocessor list and security documentation, visit the Trust Center.

Security testing

We conduct third-party penetration tests as part of our security program. Executive summaries are available under NDA on request.