Extract invoice PDFs, scans and photos to JSON and CSV with line items. Re-checks line math, tax and totals and flags errors. $0.05 per invoice.
A batch invoice processing Actor that extracts structured invoice data from PDFs and image files (PDF, JPG, PNG, WEBP, GIF, TIFF, AVIF) using OCR, recomputes invoice arithmetic, cross-checks extracted amounts against recognized document text, enforces configurable business rules (currency, allowed VAT rates, duplicate detection) and emits validated, spreadsheet-ready structured outputs (per-invoice JSON with invoice metadata, vendor/customer, dates, currency, line items, subtotals, tax breakdowns, totals and validation checks; plus CSVs: one-row-per-invoice and one-row-per-line-item). Core mechanisms: OCR text recognition, parsed invoice field extraction, numeric recomputation (quantity × unit price, line sums, tax calculations, subtotal/total/amount-due), consistency checks and rule-based flagging with severity levels, and configurable tolerance for numeric comparisons to allow for rounding.
What a run is charged for
| Event | What it covers | Price |
|---|
| Processed invoice main |
One invoice (up to 5 pages) read, extracted to JSON/CSV and validated. Failed invoices are free. |
$0.05 |
A run will not start unless its budget covers at least $1.00.
Convert any bank statement PDF, digital or scanned, to categorized transactions and CSV. Every row reconciled against printed balances. $0.015 per page.
A converter that turns bank statement PDFs (digital or scanned) into structured, categorized transaction data and a statement-level summary. It performs OCR/text extraction and parsing to produce one transaction row per printed entry (transaction date, merchant/description, counterparty, signed amount with debit/credit direction, and running balance), runs per-row and per-page balance reconciliation by comparing printed running balances and by validating opening + credits − debits = closing to the cent, and flags mismatches and brought‑forward gaps. The Actor applies AI-driven categorization using a default or user-supplied chart of accounts (up to 60 categories), detects and normalizes reverse-ordered statements (newest-first → oldest-first), and emits a CSV-ready transaction list plus a summary including account metadata, period, totals per category, aggregate credits/debits, page statuses and overall reconciliation status.
What a run is charged for
| Event | What it covers | Price |
|---|
| Statement page main |
One bank statement page read, transactions extracted, categorized and balance-checked. Unreadable pages are free. |
$0.015 |
A run will not start unless its budget covers at least $0.5.
Turn meeting recordings (mp3, m4a, wav, mp4) into minutes: summary, decisions, action items with owners and due dates, transcript. $0.012 per minute.
Converts audio or video meeting recordings into structured meeting minutes using speech-to-text and NLP: produces a concise meeting summary, extracted decisions, extracted action items with resolved owners and due dates plus the exact quote and timestamp where each item was mentioned, identified open questions with who raised them and when, speaker-labelled time-stamped transcripts (text and structured JSON), topic/sectioned notes with timestamps, and metadata such as detected language and recording duration. Technical capabilities include automatic language detection, optional minutes generation in a different language, speaker diarization and attendee matching, timestamped quote extraction, relative-date resolution when a meeting date is provided, filler-word removal, and output formatted as arrays per logical section plus a Markdown minutes document for downstream automation.
What a run is charged for
| Event | What it covers | Price |
|---|
| Audio minute main |
One started minute of the processed recording (minutes, action items and transcript) |
$0.012 |
A run will not start unless its budget covers at least $1.50.
Turn your crawled help center or docs into a chatbot knowledge pack: clean embedded chunks, a sourced FAQ per topic, a topic map and a grounded system prompt.
Converts crawled website content (help centers, docs, product pages) into a structured, retrieval-ready knowledge pack for support chatbots: it cleans and de-duplicates pages, splits content into heading-aware text chunks (~500 tokens, with overlap), generates embedding vectors (text-embedding-3-small; configurable dimensions ~256–1536), clusters pages into topical groups, synthesizes a FAQ per topic with source citations and review flags for numeric claims, produces a topic map with summaries and sample questions, and emits a grounded system prompt for chatbots with strict rules for sourcing, quoting numbers, and escalation. Outputs include heading-preserving chunk files with embeddings, FAQ files in text/MD, a topic map, and a system prompt; processing steps emphasize cleaning, deduplication, chunking by headings, embedding, clustering, and FAQ extraction from central passages.
What a run is charged for
| Event | What it covers | Price |
|---|
| Processed page main |
One unique page cleaned, chunked and embedded |
$0.003 |
| FAQ item delivered |
One FAQ question and answer with source page links |
$0.002 |
| Knowledge pack built |
Topic map, FAQ files and a grounded support-bot system prompt, once per run |
$0.01 |
A run will not start unless its budget covers at least $0.1.