Every document read, checked and routed. A person, only when it matters.
Tessera is an AI agent platform for invoices, claims and contracts. It extracts each field with a confidence score, holds the uncertain ones for human review, and sends the result where it belongs.
- 1Read
- 2Check
- 3Review
- 4Route
Source document
Extracted fields
- Supplier Halvorsen Supply Co. 0.99
- Invoice no. INV-20418 0.99
- Issued 2026-10-02 0.98
- PO reference PO-77310 0.97
- Total due 12,480.00 USD 0.99
- Bank account ...4417 0.621.00
- PO-77310 found. Total within 2% of the order.
- No duplicate in the last 90 days.
Bank account scored 0.62: the 15th digit is smudged on the page.
Result
{ "document": "invoice", "supplier": "Halvorsen Supply Co.", "invoice_no": "INV-20418", "issued": "2026-10-02", "po": "PO-77310", "total": 12480.00, "currency": "USD", "bank_account": "...4417", "reviewed": ["bank_account"], "route": "ap.approvals" }
ap.approvals
Rule: PO matched, total under 25,000 USD. Ledger draft created, original attached.
Playback is sped up. A real review usually takes a person under a minute.
Three steps from inbox to ledger
No model training, no template zones. You describe what you need, and Tessera shows its work.
-
01
Describe the fields
Name the fields you want, their types and the checks that matter: match this PO, compare this account to the vendor master. Ten sample documents are enough to start.
-
02
Read and check
The agent reads each page, scores every field, and runs your checks against your own data. Scores are calibrated nightly from reviewer corrections, so 0.90 means nine in ten.
-
03
Review and route
Uncertain fields wait in a review lane with the page region beside them. Once confirmed, rules decide the destination: ledger, case system, approver, or back to the sender.
Built for documents that move money or carry risk
Four workflows we configure most often. Each has a schema, rules and a review lane tuned to it.
- 01
Accounts payable
Read every supplier invoice, match it to the purchase order, and send only the doubtful ones to a person. - 02
Insurance claims
Turn a claim form, photos and a repair estimate into a structured file an adjuster can act on in minutes. - 03
Contract review
Pull the terms that matter out of agreements and amendments, and flag the clauses that differ from your playbook. - 04
Customer onboarding
Check identity forms, proof of address and business registrations against the application before a person opens the file.
Connects to the systems you already run
Eight kinds of connection, each one configured per workflow. No vendor lock-in on either side.
-
ERP and accounting
Post approved invoices and purchase matches straight into the ledger.
General ledger, purchase orders, vendor master -
Email and inboxes
A dedicated address per workflow. Attachments become documents, the thread stays linked.
Shared mailboxes, forwarding rules -
Cloud storage
Watch a bucket or a folder, write results and originals back where your team already looks.
Object storage, shared drives, SFTP -
Databases and warehouses
Append structured results to the tables your analysts query.
SQL databases, columnar warehouses -
Case and ticketing
Open or update a case with the extracted fields and a link to the reviewed document.
Claims systems, helpdesks, CRMs -
E-signature and contracts
Read executed agreements and push key terms into your contract repository.
Signature platforms, clause libraries -
Webhooks
Signed events for every state change: received, needs review, routed, failed.
Any HTTPS endpoint -
Queues and automation
Fan results out to message queues or no-code automation tools.
Message brokers, workflow runners
Your documents stay yours
Documents are some of the most sensitive data a company holds. Tessera's defaults reflect that.
Encrypted, regional, deletable
Encryption in transit and at rest, a US or EU region of your choice, and retention from zero days to a year.
No silent access
Our staff see a document only through a time-boxed grant you approve. Every view lands in your audit log.
No training on your data
Your documents and corrections tune your workflows only. They never improve anyone else's.
Pricing that follows volume
Pay for documents, not seats you do not need. Review is included, however many people do it.
-
Starter
One workflow, one reviewer, up to a thousand documents a month.
$79per month, 1,000 documents -
Team
Several workflows, shared review queues, and the controls finance and legal teams ask for.
$449per month, 10,000 documents -
Scale
High volume, private regions and a named engineer who knows your document types.
Customfrom 50,000 documents
Questions people ask first
What kinds of documents does Tessera read?
Invoices, credit notes, purchase orders, insurance claim forms, contracts and amendments, and similar semi-structured business documents. Scanned pages, photos and PDFs with a text layer all work. You describe the fields you want in a schema, and the agent reads each document against it.
What happens when the agent is not sure?
Every extracted field carries a confidence score. Fields under your threshold are held in a review lane, shown to a person next to the page region they came from, and released once confirmed. Corrections are logged and used to tune your workflow, never anyone else's.
Do I need to train a model on my documents?
No. You write a schema and a handful of rules, upload ten to twenty sample documents, and read the results. Most teams are routing live documents inside two weeks. You can add examples later to sharpen a field that keeps landing in review.
How are documents counted?
One document is one file you submit, up to 50 pages. A longer file counts as one extra document per additional 50 pages. Re-running the same document after a schema change within 24 hours is free. Documents sent to review count once, however many reviewers touch them.
Is my data used to train models?
No. Customer documents and corrections are used only to run and tune your own workflows. They are never used to train or improve models for other customers, and they are deleted on the retention schedule you set.
Where is my data stored, and for how long?
In the region you pick at sign-up, United States or European Union. Default retention is 30 days after processing, configurable from zero (documents are discarded once routed) to 365 days. Extracted results and the audit log follow their own retention settings.
Bring ten of your hardest documents.
We are admitting teams from the waitlist in small batches. Tell us what you process and we will set up a sandbox with you.