Stepwork logoStepwork logo
How it works

Procedures

  • OverviewHow a Procedure runs the whole process
  • Extractions deep diveRead tables and documents, use them anywhere
Deterministic AgentsMarketplace

By department

  • ITAccounts, access, helpdesk requests
  • Security & ComplianceAccess reviews, evidence, audits
  • Finance & AccountingInvoices, billing, month-end close
  • Sales & RevOpsCRM to ERP, orders, quotes
  • OperationsOrders, deliveries, dispatch
  • HR & PeopleOnboarding and offboarding
See every use case

By industry

  • HealthcareReferrals, eligibility checks, EHR entry
  • 3PL and warehousingPDF orders, WMS entry, client billing
  • Freight and truckingRate confirmations, carrier checks, TMS
  • Fuel and convenience storesStore staff, POS IDs, back office
  • ConstructionJob setup, subcontractor insurance
  • Manufacturing and distributionPO entry, supplier dates, legacy ERPs
  • Accounting firmsMonth-end across every client file
  • Field services and tradesService calls to jobs to invoices
  • MSPsNew users and license true-ups
See every industry
Security
Log inBook a Demo
How it works

Procedures

  • OverviewHow a Procedure runs the whole process
  • Extractions deep diveRead tables and documents, use them anywhere
Deterministic AgentsMarketplace

By department

  • ITAccounts, access, helpdesk requests
  • Security & ComplianceAccess reviews, evidence, audits
  • Finance & AccountingInvoices, billing, month-end close
  • Sales & RevOpsCRM to ERP, orders, quotes
  • OperationsOrders, deliveries, dispatch
  • HR & PeopleOnboarding and offboarding
See every use case

By industry

  • HealthcareReferrals, eligibility checks, EHR entry
  • 3PL and warehousingPDF orders, WMS entry, client billing
  • Freight and truckingRate confirmations, carrier checks, TMS
  • Fuel and convenience storesStore staff, POS IDs, back office
  • ConstructionJob setup, subcontractor insurance
  • Manufacturing and distributionPO entry, supplier dates, legacy ERPs
  • Accounting firmsMonth-end across every client file
  • Field services and tradesService calls to jobs to invoices
  • MSPsNew users and license true-ups
See every industry
Security
Log inBook a Demo

Extractions

Out of the app. Into the next step.

Show Stepwork a table or list once, or name the fields to pull from a document. Each run hands them to the next step: a loop, a rule, a Slack message, or another app.

Book a DemoBrowse extraction flows
OktaPeople
Table · 18 rows from 3 pages
NameEmailStatusLast sign-in
Ana Ruizana@northwind.coActiveToday
Ben Okaforben@northwind.coSuspended94 days ago
Chloe Parkchloe@northwind.coActiveYesterday
Dev Pateldev@northwind.coActive3 days ago
Elena Rossielena@northwind.coDeactivated120 days ago
Page 1 of 3Next ›
Extraction
Okta

Extract full user directory with status

Output variables

  • name
  • profile_link
  • email
  • status
  • last_sign_in

Use it anywhere

  • SlackPost it as a table
  • LoopRun steps for each row
  • API requestSend it in the body
  • AI promptSummarize or classify

How it works

Show it once. Stepwork takes it from there.

You record an extraction by doing it once.

  1. Row
    ↑
    1

    Highlight what you need

    Open the app and point at a value, a row, or a whole table. The arrow keys widen the box, and Enter captures it.

  2. nameemailstatus
    123Next ›
    2

    Stepwork learns the shape

    It finds the rows and columns, keeps links, and follows Next buttons or endless scroll, up to 100 pages.

  3. last Tuesday↓Nov 3, 2026
    3

    Tidy it in plain words

    Ask for one date format, no duplicates, a pivot, or a new column. Dates like “last Tuesday” become real dates.

  4. nameemailstatus
    Column renamed · run flagged
    4

    The columns stay put

    Column names are fixed when you save. If a page adds, drops, or renames a column later, the run flags it.

  5. For each rowSlack tableAPI request
    5

    Every later step can use it

    Rows become a list the next steps can loop over. A single value drops straight into a field, a message, or a rule.

What it can do

From page or PDF to the next step.

Read a table or a document, clean it up, and run the next steps on every row.

Tables and lists

All the rows, not just page one.

Point at one cell in any app, then widen to the whole table. Stepwork works out the columns in grids, lists, and cards, leaves out pop-ups and empty rows, and follows every page into one table.

What it reads

  • Tables, grids, lists, and repeating cards
  • Clicks Next or scrolls to load more, up to 100 pages
  • Leaves out pop-ups, hidden text, and empty rows
  • Keeps the links in each row as their own column
SalesforceOpportunities
OpportunitiesAll Open · 214 items
Search this list…Sorted by Close Date
Opportunity NameAccount NameStageAmount
Acme — Renewal FY27Acme CorpNegotiation$84,000.00
Globex — ExpansionGlobexProposal$126,500.00
Initech — PlatformInitechDiscovery$42,000.00
Umbrella — SeatsUmbrella HealthNegotiation$18,750.00
Hooli — PilotHooliQualification$9,900.00
Vandelay — RenewalVandelay ImportsProposal$61,200.00
6 of 214 itemsScroll to load more
Extraction
Salesforce

Extract open opportunities

Opportunities · All Open

Output variables

  • opportunity
  • opportunity_link
  • account
  • stage
  • amount
More pages foundLinks kept as their own column
THE PINE TAVERNmerchant · The Pine Tavern
412 Pine St · Seattle, WA
Server: AnaTable 12
10/28/2026date · Oct 28, 20268:42 PM
2 Ribeye96.00
1 Caesar salad14.00
1 Fries9.00
2 Sparkling water8.00
1 Espresso4.00
Subtotal131.00
Tax11.46
Tip26.00tip · $26.00
Total168.46total · $168.46
VISA **** 4821APPROVED
xA Moreno
THANK YOU
File processing

Read receipt photos

IMG_4182.jpg

Extracted variables

  • merchant
  • date
  • tip
  • total
RampCreate expense
Handwritten tip and total read

Documents and filesNew

Read any document, even the messy ones.

Add a File processing step to read receipts, contracts, tax forms, and invoices, whether they arrive as PDFs, scans, or phone photos. It reads printed and handwritten text, checks for signatures, and returns the fields you name.

What it handles

  • PDFs, scans, phone photos, Office files, and email files up to 7 MB
  • Handwriting, checkboxes, stamps, and signatures
  • Fields come back as text, numbers, dates, amounts, yes or no, or lists
  • Sorts documents by type, splits a batch scan, and fills PDF forms

Reshape

Clean it up in plain words.

Describe the change you want, or start from a template. Stepwork rebuilds the table before the next step gets it.

What you can ask for

  • One format for dates, amounts, and numbers
  • Duplicates removed and gaps filled
  • Pivots, group totals, and computed columns
  • A new column that labels or flags each row
Open invoices5 rows
vendordateamount
Acme Supply10/28/2026Oct 28, 2026$580.40$580.40
Northwind FreightOct 30, 2026Oct 30, 20261,250.00 USD$1,250.00
Globex Packaginglast TuesdayNov 3, 2026$96$96.00
Acme Supply10/28/2026Oct 28, 2026$580.40$580.40
Initech Rentals2026-11-04Nov 4, 20262200$2,200.00
Refine with AI

Apply
Extraction

Extract open invoices, cleaned up

Output variables

  • vendor
  • date
  • amount
  • SlackPost the table to #finance
  • API requestLoad rows into the warehouse
  • AI promptFlag unusual amounts
  • SummaryWeekly spend recap
4 rows · 1 duplicate removedOne clean table, four next steps
Extraction
Okta

List all users

Row by row · 12 users

Output variables

  • name
  • email
  • last_sign_in
Rule logic

Check last sign-in

Input variables

  • last_sign_in

Paths

  1. If last sign-in is over 90 days agoThen: Remove Slack seat
  2. OtherwiseSkip this user
Execution
Slack

Remove Slack seat

Input variables

  • email
12 of 12 users

Loops

Run the next steps on every row.

Leave the Extraction step on Row by row and every step after it runs once per row. Set it to Whole table when the next step needs all the rows at once.

How loops work

  • Row by row or whole table, set on the step
  • Loops inside loops, and loops over files
  • Rules send each row down its own path
  • With Allow partial extraction on, a failed row is marked and the rest keep going

Built to repeat

The same read, every run.

You set it up once. Later runs replay it.

  • Columns are fixed

    Column names are set when you save, so later steps find the same fields.

  • Learned once, then replayed

    Stepwork saves how the columns line up. When they line up cleanly, later runs follow that map instead of asking AI again.

  • Changes get flagged

    If a page adds, drops, or renames a column, the run flags it.

  • Long tables stay whole

    Long tables are stored and handed on in full, not trimmed, so the next step sees every row.

Examples

Start from a ready-made extraction.

A few from the Marketplace, grouped by team. Open one to see what it reads.

IT and access

  • Extract full user directory with statusOkta
  • Extract active users with license typesSalesforce
  • Extract device inventory with last check-inJamf

Finance and procurement

  • Extract AR aging and overdue invoicesNetSuite
  • Extract card transactions by department for closeRamp
  • Extract spend by category and supplierCoupa

Security and compliance

  • Extract failing tests and controlsVanta
  • Extract external file sharing reportGoogle Workspace
  • Extract secret scanning alertsGitHub

Sales and support

  • Extract stale opportunities reportSalesforce
  • Extract duplicate contact candidatesHubSpot
  • Extract SLA breach reportZendesk
Browse all 544 extraction flows

See extractions in a use case

  • Data warehouse extraction
  • User access reviews
  • Vanta evidence collection
  • Portal and website monitoring

FAQ

Questions about extractions

  • No. If the app shows it as text in a field, list, or table, an extraction can read it. That includes internal tools and vendor portals with no export at all.
  • Stepwork clicks Next or scrolls to load more, up to 100 pages. It joins the pages into one table and drops rows that repeat across them.
  • Yes, with the File processing step. It reads PDFs, scans, phone photos, Office files, CSV, text, HTML, XML, and email files up to 7 MB, including handwriting, checkboxes, stamps, and signatures. It returns the fields you ask for as text, numbers, dates, amounts, yes or no, or lists.
  • Column names are fixed when you save the extraction. If a later run finds a column added, dropped, or renamed, the run flags it.
  • Not for the parts it has already learned. You use AI to set up the cleanup. Stepwork then saves how the columns line up, and simple changes like filtering, sorting, grouping, and removing duplicates, as fixed steps that repeat runs replay. If a table can’t be lined up cleanly, that table falls back to AI cleanup.
  • Into any later step: a loop over every row, a rule, a Slack message as a table, an API request body, an AI prompt, a summary, or another Procedure. Teams load sheets and data warehouses through an API request.
  • Yes. Set the Extraction step to Row by row and the steps after it run once for each row. Turn on Allow partial extraction and a failed row is marked while the next row runs. The run finishes with the errors listed.

See it run on a process you already repeat

Bring one. We will show you what it looks like as a flow and what each run records.

Book a Demo
Stepwork

Deterministic agents for work that must run the same way every time.

ProcedureExtractionsContainersDeterministic AgentsMarketplaceUse CasesIndustriesSecurityClaude Co-Work AlternativeStepwork MCPAngelList Case StudyFAQ
Terms and ConditionsPrivacy PolicyData Processing AgreementSubprocessors

2261 Market Street #4481, San Francisco, CA 94114, USA

Loot Discount inc dba Stepwork

© 2026 Stepwork. All rights reserved.