Ledgr
Ledgr is the owned model core: a domain-tuned reasoning model that resolves AP/AR/close exceptions and a separate verifier model that gates every action before it touches the ledger.
View product →- Home
- Products
One agent team. Five owned, GPU-accelerated products.
Every product is a component of the same idea — autonomous finance operations — and every one runs on owned models, not rented API calls.
Composing the autonomous back office
LedgrDoculaVocoraTwinnReconnLedgrDoculaVocoraTwinnReconn
All products
Pick a workflow — or run the whole close
Filter by where it acts in the finance stack. Each links to a full product page.
Docula
Docula is the perception layer — an owned multimodal model that turns any vendor document into a schema-validated record with per-field confidence and audit-ready provenance.
View product →
Vocora
Vocora chases overdue invoices by phone with sub-300ms turn-taking — negotiating within policy, booking promises-to-pay, and writing structured outcomes back to the ERP.
View product →
Twinn
Twinn is the safe training ground — a replayable simulation of your ERP and close process where agents learn via reinforcement learning and every release is regression-tested against thousands of scenarios.
View product →
Reconn
Reconn resolves bank-to-ledger reconciliation as a GPU vector + graph problem — fuzzy-matching millions of lines across timing, partials, and FX, then proposing the journal entry.
View product →By layer
How the products stack
Reasoning & verification
Ledgr — the owned brain and independent verifier that decides and self-checks every finance action.
Intelligence
Ledgr — the reasoning and verifier models that decide and check.
Perception
Docula — owned document extraction feeding every workflow.
By workflow
Where they act
Accounts receivable
Vocora runs collections calls and captures promises-to-pay.
Reconciliation
Reconn matches millions of bank and ledger lines in minutes.
Safety & training
Twinn validates every agent in simulation before it goes live.
Why owned models
The moat is in the metal
Defensible
Fine-tuned on a proprietary close-cycle corpus competitors can't buy.
GPU-essential
Training, verification, vision, voice, simulation — all need accelerated compute.
Compounding
Every customer's corrections make the next close more accurate.
Proof
Benchmarked, not asserted
0%
STP on the beachhead
0%
Verifier catch rate
0ms
Voice turn latency
0
Lines reconciled / run
We started with one product and expanded to four inside a quarter. Same buyer, same platform, more of the close automated each month.
VP Finance, Brightwave
Land and expand
Begin with the workflow that hurts most. Add the rest as trust compounds — net revenue retention above 140%.
See pricing →Getting started
From pick to production
Pick a workflow
Choose the product that maps to your biggest bottleneck.
Run it in shadow
Compare agent output to your team on a recent period.
Expand the fleet
Add products as trust compounds across the close.
Product FAQ
Choosing between products
Most teams start with Ledgr + Docula on AP 3-way match — the highest-volume, highest-error workflow.
Yes, but they share memory, audit trail, and the verifier when run together through the Closora platform.
Yes — each runs owned or accelerated models that genuinely require GPUs in production.
Ready when you are
Find your first agent.
Tell us your biggest finance bottleneck and we'll show the product that fixes it.