When your AI acts inside
a financial system —
who authorized it?
"When your AI recommends or executes an accounting action, how do you prove who authorized it and what evidence it relied on?"
reht is the verified execution runtime that lets AI tools act around financial systems without giving them direct trust, keys or uncontrolled authority.
Talk to usWhere reht sits
Between your AI and the ledger.
reht does not replace your AI model or your ERP integration. It sits in the execution path — between the action your AI wants to take and the financial system it wants to touch.
The gap
Your AI can recommend. It cannot yet prove.
AI tools building on accounting APIs already handle OCR, booking suggestions and workflow automation. What they generally do not yet have:
Agent identity
No persistent, verifiable identity for the AI agent that acted.
Runtime policy enforcement
No server-side policy outside the LLM that blocks prohibited actions before they execute.
AI-specific approval flows
No structured human approval gate tied to the AI's action, evidence and scope.
AI receipts
No signed, immutable receipt proving what the AI did, what evidence it used and who approved.
Work contracts for AI agents
No defined, locked scope of what the agent may do before it starts a task.
Settlement / payout control
No mechanism to count value only on verified, approved outcomes — not on activity.
These gaps are not a problem today. They become a problem when an AI recommends a ledger entry, triggers a payment or acts without a traceable authorization decision.
What reht adds
The execution layer your AI is missing.
Agent Identity
A defined, persistent identity for each agent task — who acted, when, under which contract.
Work Contract
Locked scope before the task starts — what tools the agent may use, what it may not touch.
Policy Gate
Server-side policy check on every action, independent of the LLM. Not a prompt. A runtime rule.
Prohibited Action Blocking
Autonomous payment, ledger write and external send blocked at the infrastructure level.
Buyer / Human Review Gate
Structured approval flow — human decision captured with timestamp, evidence and action scope.
Receipt Ledger
Immutable, signable receipt per approved action. Audit trail readable by accountant or regulator.
Accepted Value Accounting
Value calculated on approved outcomes only — not on token spend or model activity.
Payout Conditions
Payment to agent or partner triggered only by verified, accepted work — not by execution.
Who we talk to first
AI and integration companies building on accounting APIs.
Not the ERP vendors. The companies moving fastest — building AI features on top of Fiken, Tripletex and PowerOffice APIs. They have customer pull and less governance depth.
Higher agent autonomy on invoice processing creates audit, approval and liability needs as workflows scale.
Risk scoring, MVA coding and forecast suggestions that influence decisions need a verified evidence trail.
LLM-to-accounting-API integrations need runtime controls before the LLM can execute against live data.
Agent-like interfaces connecting LLMs to Fiken or similar APIs are exactly the execution surface reht secures.
Connecting AI and accounting platforms across customers means governance becomes a distribution problem.
A unified API layer is a natural insertion point for reht — one integration, governance across all connected ERP systems.
ERP vendors (Visma, Tripletex, PowerOffice) are potential later partners or acquirers — after pilots exist, ARR is visible and integration risk is known.
Integration
One integration. Runtime governance across your AI stack.
reht wraps your existing AI workflow. No model changes, no ERP modifications.
Define a work contract
Specify what actions the agent may take, what evidence is required and what human approval triggers. Takes minutes to configure per workflow type.
Route actions through the policy gate
Before your AI touches any accounting API, the action request passes through reht's policy check. Prohibited actions are blocked. Permitted actions continue. Approval-required actions wait.
Capture human approval
The accountant or user sees a structured review — action, evidence, confidence, impact. They approve or reject. The decision is timestamped and signed.
Write the receipt
Every approved action gets an immutable receipt: agent identity, work contract, evidence used, human approval, outcome. Readable by the customer, their accountant or a regulator.
POST /reht/gate
{
"agent_id": "semine-invoice-agent-v2",
"contract_id": "ap-automation-no-2026",
"action": "POST /fiken/invoices/approve",
"evidence": { "ocr_confidence": 0.94, "supplier_known": true },
"amount_nok": 12400
}
{
"decision": "APPROVE_REQUIRED",
"reason": "amount exceeds auto-approve threshold (10 000 NOK)",
"review_url": "https://reht.io/review/TXN-2026-0042",
"receipt_pending": true
}
Commercial model
Price by verified workflow. Not by tokens.
No pricing by model calls or prompt tokens. You pay for verified executions, active customers or platform volume — whichever fits your billing model.
Integration pilot
Proof of fit
Free
Connect one workflow. Run 50 policy checks. See receipts.
Per customer / month
SaaS / platform model
NOK 50–200
Per active end-customer per month. You embed, we bill you wholesale.
Per execution
Usage model
NOK 1–5 / action
Per verified workflow execution. Scales with your volume.
OEM license
White-label runtime
Custom
Embed the runtime under your brand. Annual license + implementation fee.
Revenue share available if your billing model requires it. We don't insist on structure — we insist on verified value.
One question before we go further.
20-minute call. Bring the question. We'll bring the answer.