LinkedIn Intelligence follows the same Zero Data Retention (ZDR) policy as the Decision Rehearsal pipeline.
During a Function Execution
- Your input (objective, uploaded files, pasted text) is read into a memory buffer on the Cloudflare Worker
- The knowledge base (category JSON files) is retrieved from R2 — this is MAIA's data, not yours
- A Claude API call is constructed with your input, knowledge base context, and behavior instructions
- The response is returned to you as a cited analysis
- Your input buffer is released — uploaded files, pasted text, and objective are purged
After Session Close
- Your uploaded files: purged
- Your pasted draft content: purged
- Your typed objective: purged
- Your analytics data: purged
- MAIA's generated analysis: delivered to you, then purged
- The knowledge base: remains (MAIA's product data)
- Your pack balance: updated in KV (account state)
Iterative Sessions (Campaign Architect)
The Architect My LinkedIn Campaign function supports iterative refinement:
- First session: Provide context, MAIA generates plan, you download it, MAIA purges
- Between sessions: You save the plan in your own storage — MAIA has no copy
- Return session: Upload your saved plan plus new context, MAIA produces an update, you download, MAIA purges again
The iterative capability comes from you bringing the previous output back — not from MAIA retaining it.
DIAA Protection
The same Data Ingestion Audit Agent that protects the Decision Rehearsal pipeline scans data uploaded to LinkedIn Intelligence functions. Flagged content is discarded immediately. The audit log records the event but not the content. content_logged is hardcoded to false.