Documents¶
Documents in Recall are PDFs you upload to a matter. Court opinions, briefs, motions, statutes, law reviews, treatises, transcripts, records on appeal — anything in PDF form that you want to search, cite, or read.
Once uploaded, Recall extracts text, detects citations, and indexes the contents so you can search across them.
Uploading¶
With a matter selected:
- Drag and drop PDFs onto the document list, or
- Use the upload button in the document list to pick files.
Uploads accept .pdf only. You can upload multiple files at once; each one kicks off its own processing pipeline.
Every uploaded PDF goes through three stages in the background:
- Text extraction. The raw text is pulled out of the PDF. This runs quickly — usually seconds. The document is readable in the PDF viewer as soon as the file lands; extraction just makes the text searchable.
- Semantic indexing. Text is chunked and embedded. This takes longer and runs against whichever embedder you've selected. Keyword search works before semantic indexing finishes; semantic search starts returning results once it's complete.
- In-document citation extraction. After indexing, Recall scans for citations inside the document (full case, short form, Id./Ibid., supra, law, journal). These feed the Table of Authorities and the in-document citation views.
Stage 1 and stage 3 don't depend on your embedder choice. Stage 2 does — change embedders and you'll be prompted to rebuild the whole index.
Status indicators¶
Each document row shows its current processing state:
| State | Meaning |
|---|---|
| Extracting | Pulling text out of the PDF |
| Indexing | Creating embeddings for semantic search |
| Ready | Extraction and indexing both complete; document is fully searchable |
| Not readable | The PDF has no extractable text (scanned image without OCR) — searchable only by document metadata |
If a PDF is scanned and has no embedded text, Recall tags it not_readable. You can still open and read it, but full-text and semantic search won't find anything inside. Run OCR on the PDF externally (Acrobat, Preview, etc.) and re-upload.
Document kinds¶
Every document has a kind, inferred at upload time. The kind drives automatic citation formatting — Recall knows to pull reporter/volume/page from a case opinion and author/journal from a law review.
| Kind | Typical source |
|---|---|
courtOpinion |
Slip opinion, PACER PDF, ordinary reported case |
westlawOpinion |
Westlaw-formatted opinion |
lexisOpinion |
Lexis-formatted opinion |
statute |
Code section, enacted law |
lawReview |
Academic journal article |
bookOrTreatise |
Casebook or treatise |
recordOnAppeal |
Motion, brief, declaration from the record |
transcript |
Deposition or hearing transcript |
briefOrMotion |
Brief or motion as a standalone document |
secondarySource |
Other secondary authority |
other |
Catch-all |
Recall's auto-detection isn't perfect. You can override the kind on any document (see Editing metadata); changing the kind re-runs citation auto-extraction against the new type.
Metadata and citation fields¶
Alongside the kind, Recall extracts structured citation fields from the first page of the PDF. For a case opinion it grabs the caption, volume, reporter, first page, court, year, and docket/WL number. For a law review it grabs author, article title, journal, volume, page, year. Each kind has its own field set — the data model uses nullable cite* fields so unused ones stay empty.
These fields feed:
- The formatted citation on every pin from that document
- The Table of Authorities
- Context-aware citations
- Any Clerk feature that needs to name the source
Editing metadata¶
Open a document's metadata editor from its row to change:
- Name / stored filename (display only; the file on disk stays put)
- Kind (re-runs citation extraction)
- Citation fields (case name, reporter, volume, page, court, year, etc.)
- Categories (free-text tags — see below)
- Readability (
readable/not_readable)
Changes save immediately.
Categories¶
Categories are free-text tags you attach to a document for grouping and filtering. A single document can have any number of categories. Categories apply within a single matter; they aren't shared across matters.
Examples of useful category schemes: liability, damages, procedure, discovery. Or case, statute, secondary. Or anything else.
In Research mode you can filter search results by category. Pins also carry a category (single-valued, unlike documents), and can be filtered independently.
Viewing and reading¶
Click a document in the list to open it in the PDF viewer. The viewer supports:
- Scrolling and zoom
- Text selection
- Search within the document (
Ctrl/Cmd+Fin the viewer) - Jumping to a pin, which scrolls to the pinned excerpt and highlights it
Select text in the PDF to enable:
- Create pin — saves the selection as a pin with inferred pinpoint and page number.
- Send to Clerk Edit — uses the selection as context for a rewrite.
Removing a document¶
Use the document's context menu and choose Delete. That removes the document record, its extracted text, its chunks, its citations, and its PDF file on disk. Pins that referenced it aren't auto-deleted; they become orphaned (the pin remains but points to nothing).
Deletion is permanent.
Re-indexing¶
Changing the selected embedder in Settings prompts a rebuild of the whole index. You can also re-trigger indexing on a single document if something goes wrong (e.g. the indexing status got stuck) — right-click the document and choose the re-index option. Recall re-runs the chunking and embedding against the current embedder.
What Recall doesn't do¶
- No OCR. Recall doesn't run OCR on scanned PDFs. If a PDF is image-only, run OCR externally before uploading.
- No non-PDF formats. Word documents, text files, HTML pages — not supported for upload. Convert to PDF first.
- No cross-matter search. Search is scoped to the current matter.