Tosven is in pilot. Early access conversations are open for the first companies. Apply

Governed AI agents

Agents do the routine work. People sign the decisions.

Each agent in Tosven has a job description, a human manager and an autonomy level you set. It prepares the work, shows its reason and evidence, and nothing runs unless the approval matches the exact data a person saw.

L0 to L3
autonomy levels, set per agent and per type of action
30 days
of past work replayed before a level is raised
72 h
cooling period before a higher level takes effect
2
agents in production today, four more on the way

A roster, not a chatbot

Read the documentation

Most AI in business software is a text box. Tosven treats agents as staff. Each one sits on the same roster as your people, with a job description, a named human manager, a permission boundary and an autonomy level. An agent can never see or do more than its manager can. Its accuracy, its cost and every mistake it has made are open on the team screen.

The work arrives as decision cards. A card states what the agent proposes, why, which records it rests on and how confident it is. Your approval is bound to the exact content you looked at. If the invoice, the balance or your own permission changed in the meantime, nothing runs and the card comes back. Every decision is signed into the audit log, by a person or by a published rule.

Agents

The agent roster, as the team screen shows it

Six agents on one roster. Two are in production in the pilot today. Four are shown as designed and are on the way. Each card names the agent, its manager and its current level.

Sample data · agent roster

What every agent carries, by construction

01

Job description and manager

Each agent has a written scope and a named human manager. Its permissions are the overlap of its own grants and the manager's, checked again at the moment the work runs.

02

Autonomy levels L0 to L3

L0 suggests only. L1 acts on exact matches. L2 acts below a threshold you set. L3 has full authority in a defined area. Amount and daily limits apply per type of action.

03

Decision cards

Every suggestion is a card with the reason, the evidence and the confidence, including what that confidence is based on. A card that lacks any of the three cannot be created.

04

Approval bound to content

Your signature covers the exact content and source versions you saw. If either changed before execution, the card turns stale, no command runs and it returns to the queue.

05

Signed audit log

Proposer, decider and executor are three separate records. Decisions are chained per company and cannot be edited or deleted, whether a person or a published rule made them.

06

30 day replay

Before a level is raised, the last 30 days are replayed at the new level. The raise also needs a second approver and a 72 hour cooling period. Lowering a level is instant.

07

Accuracy from outcomes

Accuracy is measured from results verified at the source, such as an invoice that was settled or an entry that was never corrected. Clicking approve does not count as correct.

08

Budgets and provider allowlist

Each run has a step, time and cost budget enforced in SQL. Model providers are limited by an allowlist that changes only with a product release, and every call is logged.

09

Honest abstention

When an agent lacks history, evidence or a fresh source, it says so with a reason code and the next step for a person. It does not guess.

In practice

Three cases, step by step

01

A purchase invoice is coded and approved

A freight invoice from Baltic Freight OÜ arrives and the Ledger agent prepares the entry for the accountant.

  1. 07:41Ledger agentReads BIL-2026-0733 from Baltic Freight OÜ, 3.720,00 €, and finds 11 earlier invoices coded to freight.
  2. 07:41Ledger agentOpens a card: freight account, VAT 25,5 %, confidence 91 %. Evidence: the 11 posted invoices.
  3. 08:12Sara NyströmOpens the card, checks the evidence links and approves. Her signature covers that exact content.
  4. 08:12Ledger agentRe-checks the source versions and Sara's permission, then posts JRN-2026-01284. One card, one effect.
  5. LaterLedger agentCounts the card as correct only when the period locks without the entry being corrected.
02

The data changed, so nothing ran

A reminder for Nordlys AB is approved in the morning, and the payment lands before the letter leaves.

  1. 08:02Collections agentProposes reminder 2 of 3 for INV-2026-0418. Open balance 48.000,00 €, 16 days overdue.
  2. 08:20Elina KoskinenReads the full letter on the card and approves it.
  3. 08:31Sara NyströmApproves the bank match of PAY-2026-0391 to INV-2026-0418. The open balance is now zero.
  4. 08:40Collections agentRe-reads the balance before sending. The source changed, so the card turns stale. No letter is sent.
  5. 08:40Collections agentThe stale execution is written to the audit log. The reminder ladder for this invoice stops.
03

Raising the Ledger agent from L1 to L2

Elina Koskinen wants small invoices from known suppliers posted under a published rule, without a click.

  1. 02.11.Elina KoskinenRequests L2 for invoice coding: known suppliers only, limit 500,00 € per invoice.
  2. 02.11.Ledger agentReplays the last 30 days at L2: 412 cards, 0 critical errors, 3 would still go to a person.
  3. 03.11.Mikko LehtonenSigns as second approver. The 72 hour cooling period starts.
  4. 06.11.Ledger agentL2 takes effect. Each automatic posting is signed in the name of the published rule.
  5. 19.11.Elina KoskinenFinds one wrong VAT box in the error log and lowers the level to L1. It applies at once.

Sample data

Ledger agent

The Ledger agent prepares the books, with its reasons

In production today. It codes incoming purchase invoices, proposes matches between bank lines and open items, and watches suspense balances and period-close tasks. It learns a coding rule per supplier only from invoices that people coded, tests the rule on invoices it has not seen and runs it in shadow before anyone can publish it.

  • Proposes account, VAT code and period for each purchase invoice
  • Bank match candidates with reference and amount as evidence
  • Learned rules are published by a person and kept on the record
  • No entry reaches the ledger without a person or a published rule
Collections agent

The Collections agent follows every overdue invoice

In production today. It watches open receivables and proposes payment reminders on a ladder of up to three steps, with minimum intervals per country. It is a rule engine and makes no model calls. It charges no fees and no interest. A pause, a dispute or a payment promise stops it at once, and a partial payment always goes to a person.

  • The card shows the full letter in the customer's language
  • The balance is re-read before every send, so the amount is current
  • A daily send limit per company is enforced in the database
  • Final reminders and first letters to a customer stay with a person
Controls

Limits the product enforces, whatever the settings say

Some ceilings are product law. No agent executes a payment. Tax filings and ledger postings stay under human approval unless a documented rule has been published. Every payment, any new or changed IBAN and any entry above 10.000,00 € need two different people. A kill switch stops a single run, one agent or a whole company.

  • The payment ceiling is zero: agents propose, people pay
  • Two signatures can never come from the same person
  • A changed model or rule version resets earned autonomy to L0
  • A timeout is never a yes: unanswered cards expire
Next agents

Four more agents, each with its walls drawn first

Treasury, Sales, Purchasing and Compliance agents are designed and are not yet in the pilot. Treasury will propose payment runs from open payables and the cash forecast. Sales will raise follow-up cards from deal signals. Purchasing will propose orders from lead times and minimum quantities. Compliance will watch VAT deadlines, filings and period locks.

  • Treasury agent: prepares a payment file, a person uploads it
  • Sales agent: deal facts only, no scores for people
  • Purchasing agent: starts at L0, raised only after a replay
  • Compliance agent: flags and reminds, never files a return itself

In the field · Property management company

Forty buildings, one accountant, and the routine prepared by 07:00

A property management company keeps the books for dozens of housing companies and commercial buildings. Each building receives the same kinds of invoices every month: energy, cleaning, lift service, waste. Each has tenants and owners who pay late in the same patterns. The work is repetitive, the volume is high, and a wrong VAT code or a reminder sent after a payment becomes a complaint from a board chairman.

In Tosven the Ledger agent codes the recurring supplier invoices for each building and brings the unusual ones forward with a reason. The Collections agent keeps the reminder ladder for every overdue rent or service charge and stops when a tenant disputes the amount or promises a date. The property accountant approves a short queue in the morning. The manager of each agent sees accuracy, cost and errors on the team screen.

With separate tools

  • Recurring invoices are coded by hand, building by building
  • Reminders depend on a spreadsheet and on someone's calendar
  • A reminder goes out the day after the tenant paid
  • Nobody can show who approved an entry, or on what basis

In Tosven

  • Recurring invoices arrive coded, with earlier invoices as evidence
  • The reminder ladder runs per invoice and stops on a dispute or a promise
  • The balance is re-read before sending, so a paid invoice gets no letter
  • Every entry shows the card, the signature and the rule behind it

A working day

Sample data
  1. 07:00The Ledger agent has prepared coding cards for 38 supplier invoices across 12 buildings. 4 are flagged for review.
  2. 08:00Sara Nyström works through the queue. Two cards are corrected, and each correction is recorded.
  3. 08:30The Collections agent proposes 9 reminders. Elina Koskinen signs them in one gesture, each card still itemised.
  4. 13:00A tenant disputes a service charge. Laura Virtanen marks the dispute and the ladder stops on every channel.
  5. 16:00Jonas Berg opens the error log: one wrong account this month, already corrected and counted in the score.

Status

Where it stands today

Live in pilot
  • Ledger agent: invoice coding and bank match proposals
  • Collections agent: reminder ladder with pause, dispute and promise
  • Decision cards with reason, evidence, confidence and signed decisions
  • Stale check: changed data or a lost permission stops the execution
  • Autonomy levels with replay, second approver and 72 hour cooling
  • Provider allowlist, run budgets, kill switch and a read-only auditor view
On the way
  • Sales agent: daily follow-up cards from deal signals
  • Treasury agent: payment run proposals, never an executed payment
  • Purchasing agent: order proposals, once stock and purchasing arrive
  • Compliance agent: VAT deadline and filing watch
  • Agent test bench and a monthly control report for the auditor
  • Reminder fees and interest, SMS and e-invoice reminder channels

Frequently asked

Questions and answers

hello@tosven.com

Can an agent post to the ledger without a person?

Only under a rule that a person published. By default every entry an agent proposes waits for a human signature. A company can publish a documented posting rule and raise the agent to L2 for that rule, after a replay, a second approver and a cooling period. Each automatic posting is then signed in the name of that rule.

What happens if the data changes after I approve?

Nothing runs. Your approval is bound to a hash of the card content and to the versions of the records it was computed from. At execution Tosven recomputes both and asks again whether you still hold the permission. If anything differs, the card is marked stale and returns to the queue with the reason.

Can an agent make a payment?

No. The payment ceiling is zero as a product rule, not as a setting. The planned Treasury agent will propose a payment run and prepare a payment file. Two different people sign it and a person uploads it to the bank. Any new or changed IBAN also needs two signatures.

Which AI models do the agents use?

Agents may call only the model providers on an allowlist, defined per class of data. The list changes only with a product release, never from a settings screen, and every call is recorded with its cost. The Collections agent uses no model at all: its reminder ladder is a deterministic rule.

How do you measure whether an agent is right?

From outcomes verified at the source, not from approvals. A coding counts as correct when the period locks without the entry being reversed or corrected. A reminder is verified when the invoice is settled. A wrong VAT box, a wrong period and a wrong counterparty count as critical errors and are listed in the open error log.

Early access

Be among the first companies.

Tosven is in pilot. Tell us about your company and the tools you use today in a few sentences, and we will say plainly whether it is a fit.

sales@tosven.com · No card details and no commitment.