READ-PDF — extract text from a PDF and inject it into the conversation context
/read-pdf FILE [PAGES]
/read-pdf extracts text from a PDF file using the poppler utilities (pdfinfo, pdftotext, pdfimages) and injects the result into the conversation as a user-role context message. The model can then reason about the content immediately.
FILE may be an absolute path, a path relative to the workspace root, or a home-relative path beginning with ~/. Unlike /read, /read-pdf is not restricted to workspace files.
PAGES is an optional page range in the form FIRST-LAST (e.g. 40-55) or a single page number (e.g. 10). When omitted, the entire document is extracted. A hard limit of 20 pages per call applies to keep the context window manageable; specify a range if the document is larger.
Three poppler tools are used in sequence:
pdfinfo — document metadata (title, author, page count, creation date) pdftotext — text extraction preserving spatial layout (-layout flag) pdfimages — raster-image inventory used to detect diagram-only pages
Pages that have sparse text and no raster images are flagged as likely vector-diagram pages. Those pages cannot be extracted by any text tool; the output will note them so you can follow up with a vision-capable model.
The injected content is ephemeral — it is added to the current conversation and is not written to disk or stored in the RAG database. Use /rag ingest if you want to index a PDF for retrieval.
Read the first ten pages of a PDF:
harvey > /read-pdf ~/docs/oberon2.pdf 1-10Read a specific section by page range:
harvey > /read-pdf ~/docs/oberon2.pdf 49-63
harvey > Summarise the module system described in those pages.Read a short PDF (≤ 20 pages) in full:
harvey > /read-pdf project/spec.pdf/read FILE… — inject plain-text workspace files into context /rag ingest PATH — index a file into the RAG store for retrieval /status — check remaining context window space /help read /help rag