add_to_workspace
ChatGPTAdd a PDFBetter document to a workspace. pinRevision=true pins the document's current version so answers keep citing it after later edits; otherwise the workspace follows the latest version.
approve_workflow_step
ChatGPTDecides the review step a run is waiting on — only when the workflow owner turned on agent approval for that step. Otherwise it returns an error: a person must approve in the PDFBetter dashboard. Corrected values apply to extraction reviews.
ask_document
ChatGPTAnswer one question using only the document's text, with citations: each citation is a page number and a short quote verified verbatim against that page. If the document doesn't contain the answer, answered is false and nothing is invented. Uses AI credits like one chat message. Scanned pages need ocr_document first.
ask_workspace
ChatGPTAsk a question across the documents in a workspace. Returns a cited answer: every citation names the document, its version and page, with a quote verified on that page. Says so when the documents don't answer it. Uses AI credits.
cancel_workflow_run
ChatGPTStops a run that hasn't finished. Steps already completed (and their usage) are kept; remaining steps are skipped.
check_pdf_edits
ChatGPTLook for signs that a document's PDF was changed after it was created: earlier saved versions (incremental updates) and the text they changed, editing software in the metadata, date gaps, digital signatures that no longer cover the file, numbers drawn in a font/size/colour their line or column doesn't use, text covered by boxes, hidden layers, image-only pages, and (for bank statements) balances that don't reconcile. Returns signals with severity, pages and innocent explanations; it never says a document is fake or genuine. Reads the stored original; a document already edited in PDFBetter is checked as its current re-saved version, without earlier history.
compare_documents
ChatGPTCompare two documents (originalId + revisedId) or two revisions of one document. Returns FACTS — verbatim before/after excerpts with page numbers in each version, changed amounts/dates/durations — and, separately, an AI INTERPRETATION grouping the changes with plain-language explanations and a low/medium/high materiality hint. The facts are free; the interpretation costs AI credits and is skipped with a note if credits run out. The interpretation is not legal advice.
compress_document
ChatGPTMake a PDF document smaller by recompressing images (light, balanced or strong). Runs as a background job and returns a jobId; poll get_job for the compressed PDF's download link. The original document is untouched. Costs one server conversion.
convert
ChatGPTConvert a PDF document to Word (docx), Excel (xlsx), PowerPoint (pptx), CSV, plain text (txt) or images (png/jpg), or convert uploaded Word/Excel/PowerPoint/text files to PDF (to: pdf, with uploadIds from create_upload). Runs as a background job and returns a jobId; poll get_job for the output download links. Costs one server conversion (plus OCR pages when ocr is on).
create_document_from_upload
ChatGPTTurn a PDF you uploaded with create_upload (after the PUT finished) into a document in the user's account. Fails if the upload isn't complete or the file isn't a valid PDF. Returns the new document id.
create_document_from_url
ChatGPTDownload a PDF from a public https URL into the user's account as a new document. Private/internal addresses are refused and redirects aren't followed (pass the final URL). The file must be a real PDF within the plan's size limit. Returns the new document id.
create_from_template
ChatGPTStart a new, private document from a template: a library slug or the id of one of the user's own templates (from list_templates). The template itself is never changed. Returns the new document id, ready for list_fields and fill_fields.
create_upload
ChatGPTGet a presigned PUT URL for uploading a local file (use this for files you can't pass by URL). PUT the exact bytes to url with the returned headers within 10 minutes, then call create_document_from_upload (PDFs) or convert with uploadIds (Word, Excel, PowerPoint, text → PDF).
create_workspace
ChatGPTCreate a document workspace. Counts toward your plan's workspace limit.
export_document
ChatGPTRender the document's current revision (with filled fields, edits and annotations) to a PDF and return a short-lived download link. Counts as one export_document on the user's plan the first time this content is exported; re-exporting unchanged content is free. flatten burns fields into the page so they can't be edited.
fill_fields
ChatGPTFill a document's fields by name (from list_fields). Text fields take strings, checkboxes true/false, dropdowns and radio groups one of their options, null clears a value. All values are validated first; if any is invalid nothing is saved and the errors say why. Saves a new revision (earlier revisions stay in the document's history).
fill_from_documents
ChatGPTProposes values for a fillable form (formId) from the user's own source documents (sourceIds: W-2, pay stubs, IDs, statements, letters) and optionally their saved 'My info' profile. Every value cites its source document, page and a verbatim quote verified against that page's text (or the My info entry); anything that can't be verified stays empty with a reason, never guessed. Choice fields only take existing options. With apply: true the high-confidence values are saved as a new revision of the form (like fill_fields); values flagged for review (rewritten, combined, replacing an existing value, unconfirmed choices) are returned as needsConfirmation and never written. Otherwise nothing changes. Scanned sources: the first call lists pages without text and runs nothing; pass ocr: true to OCR them (OCR pages) or ocr: false to skip them. Costs AI credits (from 3, by length). The form needs fillable fields: if it has none, open it in the editor and detect fields first.
generate_brief
ChatGPTWrite a source-grounded brief (kind=brief) or slide outline (kind=slides) from a workspace's documents. Returns markdown with [n] citation markers, the verified citations (document, version, page, quote) and a short-lived PDF link. Uses AI credits.
get_document
ChatGPTGet one document's details: title, page count, current revision number, how many fillable fields it has, and a link to open it in the PDFBetter editor. Use before filling or transforming a document you only know by id.
get_help_article
ChatGPTThe full text of one PDFBetter help article as Markdown (steps, limits, FAQs), by slug from search_help. Cite its URL when you use it. Free: no agent_call is counted.
get_job
ChatGPTCheck a background job started by convert, compress_document or ocr_document: status (queued, running, succeeded, failed), progress percent, error message, and short-lived download links for its outputs once it succeeded. Poll every few seconds; links expire after about 10 minutes, so call again for fresh ones.
get_share_activity
ChatGPTActivity on the user's share links: views, unique visitors per day, downloads, comments, fills, wrong-password attempts and hits on expired links, ranked by most recent activity, plus recent events. Pass linkId for one link's per-day timeline. Visitors are counted with a daily-rotating hash (no IP stored, no cross-day tracking); a view is not verified identity, and link previewers can occasionally register as views.
get_workflow_run
ChatGPTReturns a run's status and per-step states (status, attempts, error, output ids such as extractionRunId or revisionId, counts, and a short excerpt for summarize steps). Extracted values are not included.
list_documents
ChatGPTList the PDF documents in the user's PDFBetter account, newest first. Use query to match titles and tags, kind to show only templates or forms, and pass nextCursor from a previous call as cursor for the next page. Returns ids you pass to every other document tool.
list_fields
ChatGPTList a document's fillable fields with their names, types, options, required flags, current values and pages, plus the document revision. Call this before fill_fields: fill by the exact name and use one of options for dropdowns and radio groups.
list_form_responses
ChatGPTList submissions to one of the user's published forms, newest first, with the submitted values keyed by field name. Responses can contain personal data: only use them for the task the user asked for. Page with cursor (nextCursor from the previous call).
list_form_submissions
ChatGPTRecent submissions to the user's published fillable forms: which form, when, edition, and the respondent email if the form asked for one. Does not return answers (use the forms tools with forms:read for response data).
list_forms
ChatGPTList the user's published fillable web forms with their status (live, paused, closed), public URL, field count and response count. Use the form id with list_form_responses to read submissions.
list_templates
ChatGPTList document templates: the user's own saved templates (type user, id = document id) and PDFBetter's library of common forms and agreements (type library, id = slug). Filter with query. Pass an id to create_from_template to start a new document from it.
list_workflows
ChatGPTLists the account's saved workflow recipes (sequential steps such as OCR → extract → review → export) with their trigger and steps.
list_workspace_outputs
ChatGPTList the saved answers, briefs, slide outlines, quizzes and audio briefs in a workspace.
list_workspaces
ChatGPTList your PDFBetter document workspaces (collections of documents you can ask across and turn into cited briefs).
merge_documents
ChatGPTCombine two or more documents into one new PDF document, in the order given (all pages of each). Form fields are kept; clashing field names get a numeric suffix. The originals are untouched. Returns the new document id and a download link.
mind_map_document
ChatGPTBuild a mind map of a document: a central topic, main branches and points, where every node below the root cites one page and a quote verified verbatim against that page (nodes that fail the check are removed and counted in dropped). Returns the tree as root {label, pages, quote, children} plus a Markdown outline. detail is compact | standard | detailed; optional focus. Uses AI credits like the app's mind map, charged only when verified nodes exist. Scanned pages need ocr_document first.
ocr_document
ChatGPTMake a scanned document searchable: adds an invisible text layer to pages without text (pages look identical) and saves it as a new revision of the same document, so summarize_document and ask_document can read it. Runs as a background job; poll get_job. Charged per page actually recognised (pages that already have text are skipped unless force is true).
rotate_pages
ChatGPTRotate pages clockwise by 90, 180 or 270 degrees (relative to their current rotation) and save the result as a new document; the original is untouched. Rotates every page unless pages is given.
run_workflow
ChatGPTStarts a saved workflow on a document and returns the run id right away. Steps run in the background one at a time; review steps pause the run until approved. Poll get_workflow_run for progress. Each step uses its own meter (OCR pages, AI credits).
search_help
ChatGPTSearch PDFBetter's product documentation (help.pdfbetter.com) for how-to answers: features, settings, plans and limits, billing, signing, forms, integrations, privacy and troubleshooting. Returns the best-matching articles with their one-paragraph answer and slug; call get_help_article for the full steps. Free: no agent_call is counted.
split_document
ChatGPTSplit a document into several new PDF documents: by page ranges (one document per range), every N pages, or one per page. The original is untouched. At most 20 parts per call. Returns the new document ids and download links.
suggest_redactions
ChatGPTSuggest sensitive text in a document for redaction review, with page numbers and boxes. Pattern categories (email, phone, iban, credit_card, us_ssn, tr_national_id, es_national_id, fr_nir) are free and checksum-validated where possible; AI categories (names, addresses, organizations, other IDs, financial, health) cost AI credits per page. Returns suggestions ONLY: it never changes the document. Suggestions can miss things. Applying redaction (which removes the content) stays a human action in the PDFBetter app's AI redaction review or editor.
summarize_document
ChatGPTSummarize a document with key points that cite the page numbers they come from, plus what the summary leaves out (unreadable scanned pages, pages beyond the plan's AI page limit). Optional focus narrows it (e.g. 'payment terms'). Uses AI credits like the editor's summary. Scanned pages need ocr_document first.