All actions

Find actions and available guidance.

Agent Research Search and read public sources to answer a question about each row.

Ask Answer a free-form question about each row's selected content and write one answer column with the selected data type.

Call an API per row Send a templated HTTP request for each row and retain its JSON object response. Requests may have external side effects; interrupted requests can be repeated on retry.

Capture page Save HTML page artifacts or extract links from selected URL cells.

Capture screenshot Render link or text URL cells in Playwright and write one PNG image cell per row.

Census Append US Census ACS demographic fields to rows with a geographic point.

Classify rows Assign labels locally or classify structured fields with a model.

Clean column Normalize values in a text-like column while preserving the source.

Clean dates Parse messy date strings into strict ISO dates.

Cluster duplicate values Group a column's duplicate values and save their reviewed canonical forms.

Combine values Rewrite explicitly grouped source values to their canonical value while preserving the source column.

Download media Download audio or video and optional subtitles, thumbnails, and metadata with yt-dlp.

Extract columns from JSON Extract dotted or indexed JSON paths into columns.

Extract faces Detect faces in each image and return pixel bounding boxes and cropped face images.

Extract fields Extract typed fields from each row, with optional source citations.

Extract metadata Extract local metadata from blob-backed image, audio, video, or file cells as one details object or useful typed fields with details.

Extract range Extract one clip per row, retaining compatible transcripts and annotations.

Extract text with regex Extract a regex match from one or more text columns.

Fetch URL Download a URL into a blob-backed file; enabled media plugins retain their native downloader.

Fill missing values Fill null or blank source cells from neighboring rows, a fixed value, or a column aggregate while preserving the source column.

Find grounded occurrences Scan every source window and create evidence-linked findings.

Find topic changes Segment each selected timestamped transcript and write complete, editable topic-section ranges.

Find visual cuts Find scene boundaries in each selected video and write editable, source-bound timestamps.

Formula Calculate a new column using Excel-style formulas and named columns.

Geocode Geocode an address column or composed address with the selected provider.

Join sheets (by key) Match exact keys from two sheets and create a joined table.

Judge results Grade an AI-generated column's values against your guidelines with a second model, writing verdict and judge-note columns plus receipt evidence for the guidelines, output roles, and model calls. The judged column is included automatically so the model sees the value under review; set include_original_prompt to add the upstream instruction that produced an AI-generated judged column.

Materialize RSS enclosures Download selected RSS enclosures into their existing media cells.

Materialize semantic join links Create a link table from a completed semantic join and its applied review decisions, without rerunning embeddings.

Named entities Extract labeled named entities with spaCy, GLiNER, or an LLM.

OCR Read text from images and documents with the selected OCR engine.

PDF tables Extract normalized table rows and source/page metadata from each blob-backed PDF.

Quick search Search once per row and answer from the result snippets, keeping the search results.

Replace by rules Apply ordered contains, exact, or regex rules to whole source values; the first match wins.

Run trusted local Python Run a trusted local Python transform over selected JSON row values, routing results to columns, named results, and receipt evidence.

Semantic join Match a source column against a target column using embeddings, with a linked results sheet.

Split into segments Split audio or video into exact clips with their transcripts and temporal annotations.

Split transcript Create transcript excerpts using complete intersecting chunks and their exact timestamp evidence.

Structured research Search and read public sources to fill typed fields for each row.

Substitute values Rewrite exact source values through an explicit mapping while preserving the source column.

Summarize grouped rows Synthesize one faithful summary per group into a new sheet.

Summarize rows Summarize each row's selected content and write one text summary column.

Table from list Expand existing list cells or a named result into a child table.

Template Build text from values in each row.

To geo point Build a geo_point from existing latitude and longitude columns. Validates ranges locally and does not call a geocoder.

To Markdown Convert documents (html, docx, pptx, xlsx, text PDFs) to markdown. Engine-selectable: local 'markitdown' (default, offline) or 'trafilatura_html' (HTML extraction), 'docling'/'chandra' (sidecar, layout-aware / experimental document VLM), or 'datalab' (hosted Marker API, pay-per-call).

Tool-assisted Extract Extract typed fields per row using selected project-local MCP tools.

Transcribe Transcribe audio and video with timestamps using the selected speech engine.

Translate rows Translate selected row content, optionally saving its detected source language.

Video frames Extract still frames from each video with timestamps and optional downscaling.