What's really inside your files

One guide per format: the file's anatomy, what breaks when a model reads it raw, and the exact element mapping the converter applies — limits included.

• Jupyter notebooks

  Cell addresses, dependency hints, execution-order warnings, base64 extraction — the hardest format, and the biggest win.

  .ipynb

• Word documents

  Styles become a real outline; tables survive as GFM; page furniture is dropped.

  .docx

• Slide decks

  Slides become an addressable outline; speaker notes surface; charts are confessed, not faked.

  .pptx

• Emails

  Decoded headers, the newest message intact, quoted reply pyramids truncated explicitly.

  .eml

• Excel workbooks

  ISO dates instead of serials, cached formula values, one section per sheet, explicit truncation.

  .xlsx · .xls

• CSV / TSV

  Sniffed delimiters, RFC-4180 quoting, typed columns, ragged rows repaired and confessed.

  .csv · .tsv

• PDF

  Per-page text with honest limits: no fake tables, scans flagged, layout loss disclosed.

  .pdf

• Webpages

  Reader-style article extraction: content kept with tables and figures, chrome discarded.

  .html

• LaTeX sources

  The structure the PDF destroys: outline, fenced math, keyed citations — no TeX engine pretensions.

  .tex

• Mailbox archives

  A whole mailbox as an addressable thread outline, quote pyramids truncated per message.

  .mbox

• JSON / JSONL

  A structure outline plus record tables — and when to leave JSON as JSON.

  .json · .jsonl

• Subtitles

  Timestamped transcripts with compact time markers; styling and numbering stripped.

  .srt · .vtt

• Markdown / plain text

  Structure QC, exact token counts, and retargeting to chat, RAG, Obsidian or archive presets.

  .md · .txt

Missing a format? Request it with a sample file — parsers here are test-driven. · Open the tool · Field manual · Context Lab