What document intelligence gives you
PDF + DOCX
Both supported
Server-side
Trusted extraction
Cached
Once, reused everywhere
RLS
Workspace-isolated
How document intelligence works
PDF + DOCX upload, server-side extraction
Upload PDFs and Word docs during intake. We extract text server-side using `pdf-parse` and `mammoth`, cache the plaintext, and never trust client-side extraction.
Feeds AI context — not a dead archive
Extracted text flows into persona generation, ICP refinement, analysis module context, and AI advisor prompts. Your old brand book actually informs the new strategy.
Workspace-isolated, RLS-enforced
Each document is scoped to your workspace. Row-level security policies prevent cross-workspace leakage. Deleting your workspace deletes the docs.
Cached on first read, fast on every reuse
Extraction runs once. Cached plaintext (`intake_documents_text`) is reused by every downstream feature, so AI calls stay fast and we don't re-parse your 60-page brand book five times.
Where extracted text shows up
Persona Studio
Brand voice, customer insights, and survey responses inform persona generation directly.
Marketing Analysis
Analysis modules use extracted text as context — they know what brand assets you actually have.
AI Advisor
Advisor cites your real documents instead of inventing details. (Eventually retrieval-backed via pgvector.)