Features — read, route, review, record
What would your month-end close look like if no field were ever retyped?
Seven capabilities, one pipeline: typed extraction, graph workflows, confidence-gated review, audit diffs, connectors, access control and run-level reliability. Each is shown below as it appears in the product.
Typed extraction: decimals, dates and enums, not text
Prebuilt models for invoices, receipts, contracts and forms return values in their native types: decimal totals, ISO 8601 dates, currency codes, enum payment terms. Line items come back as arrays, with a bounding box and a confidence score on every value.
Need a field we don’t ship, like a carrier’s fuel-surcharge code? Label it on 25 of your own documents and train a custom model against the same schema.
- Invoice, receipt, contract and form models
- Per-field confidence, 0.000–1.000
- Custom fields from your labelled samples
First station in every workflow. Its output is the schema that later conditions, rules and connectors read.
A workflow builder that branches on the data itself
Conditions read extracted fields, not file names. Send invoices over $10,000 to a named approver, drop unknown vendors into review, and post everything else straight to QuickBooks.
Every edit is a versioned draft. Run it against ten sample documents, read the diff against the live version, then publish.
- Branch on any field, vendor list or confidence value
- Approval steps with named approvers
- Test runs before publish; one-click rollback
Sits between extraction and your accounting tool, deciding who sees a document and where it lands.
Pick a ceiling
Every capability above ships on the three yearly plans.
The difference is pages per month: 500, 10,000 or unlimited. Compare the ceilings side by side.
Compare plan ceilingsConfidence thresholds that decide what a human sees
Set an auto-approve threshold per field or per document type, then add validation rules: subtotal + tax must equal total; PO number required above $5,000. A document that breaks either goes to the review queue with the failed rule and the source region attached.
Reviewers correct one field, not a whole document, and work the queue from the keyboard.
- Thresholds per field or document type
- Arithmetic, regex and required-field rules
- Assignment, J / K navigation, bulk approve
Between extraction and posting: the gate that keeps doubtful values out of your ledger.
An audit trail that shows the before and the after
Each run stores the input file, the model version, the workflow version, every field value, every edit and every approval, with timestamps in UTC. When an auditor asks why a figure changed, you open the run.
Runs are searchable by vendor, amount, reviewer and date range, and can be exported as CSV for your own records.
- Before / after diff on every edit
- Named reviewer and approver per run
- Filter by vendor, amount, user, date
Wraps every station. Nothing posts to a downstream tool without leaving an entry here.
Connectors that push typed records, not CSV attachments
Mapsy posts to QuickBooks, Xero, Salesforce, HubSpot, Slack, Google Workspace and more than 30 other apps, or straight into your database over an authenticated API. Field mapping happens once, in a panel, and each push is stamped into the run history.
- OAuth connections, revocable per workspace
- Field mapping with test push before go-live
- Webhook and REST API for everything else
The last station: after validation and approval, the record lands where your team already works.
Roles, single sign-on and approvals that follow your org chart
Give AP clerks the review queue, controllers the approval step and auditors a read-only view of the log. Orchestrate adds SAML single sign-on, so access ends in your identity provider the day someone leaves.
- Admin, Builder, Reviewer and Viewer roles
- SAML SSO on the Orchestrate plan
- Multi-step approvals with named approvers
Controls who can edit a workflow, who can approve a payment-sized invoice, and who can only look.
Throughput you can watch, and retries you don’t have to
A volume chart shows documents per day against your plan ceiling. If a downstream tool is unavailable, Mapsy holds the record, retries with backoff and posts it once the connection returns. No duplicate bills, because every push carries an idempotency key.
- Retries with exponential backoff
- Idempotency key on every push
- Alerts to Slack or PagerDuty on stalled runs
Runs underneath all stations, so a slow accounting API delays a bill, not loses it.
Security and reliability
The controls, listed plainly
What protects your documents and your ledger, written as a spec rather than a promise.
- In transit
- TLS 1.2+ on every connection, including webhooks and connector calls.
- At rest
- Documents and extracted fields encrypted with AES-256; keys rotated on a fixed schedule.
- Access
- Four roles, SAML SSO on Orchestrate, and a read-only view for auditors.
- Retention
- You set how long source files are kept; the audit log keeps values and diffs after files expire.
- Delivery
- Retries with exponential backoff and idempotency keys, so a flaky API never creates a duplicate bill.
- Model changes
- Model versions are pinned per workflow; an upgrade is a draft you test, never a silent swap.
Connector index
Forty-plus apps, sorted by what they do with the record
Next step
Seven capabilities, three ceilings. Pick the one that fits your inbox.
Every plan includes the full pipeline; the plan sets how many pages a month it reads.
Choose your plan