Supported document work
PrivateAI for Mac currently supports searchable PDFs, Markdown, plain text, HTML, JSON, CSV, XML, YAML, and common source files. Users can select files through the system picker, drag files from Finder, or include an existing supported local path in the request.
The App copies selected attachments into private managed storage under ~/.privateAI/artifacts. Model-facing document Tools receive bounded content and provenance instead of unrestricted filesystem access. Explicit path references authorize that exact file for the request, not its siblings or containing directory.
Long-document analysis
Large documents can be processed through hierarchical analysis rather than forcing the entire source into one model request. PrivateAI divides source material into bounded units, reduces intermediate summaries recursively, and saves private checkpoints under ~/.privateAI/jobs/document-summaries. Repeating the same document, goal, and model can reuse completed work.
Document privacy mode
- Document conversations expose local resource and document analysis Tools.
- Public Web, Browser, Terminal, and native service Tools are not exposed in that conversation mode.
- Document bytes and extracted text are not written into ordinary Tool transcript rows or runtime logs.
- Intermediate summaries and final answers may contain quoted or summarized source material and remain local.
Who this is for
This workflow is designed for people who want a native Mac interface around local model inference: reviewing reports, searching technical references, summarizing research, understanding code or data files, and resuming long analysis without uploading source documents to a hosted model provider.