Skip to content

AI Accountant

Intelligent accounting assistant for Norwegian companies. Supports Tripletex and Fiken.

An AI agent that connects to your accounting system (Tripletex or Fiken) and handles routine bookkeeping — monitoring compliance, processing invoices, answering employee questions, and running periodic tasks like month-end closing.

What It Does

Category Examples
Proactive Monitoring Overdue bill alerts, expense policy violations, month-end reminders
Employee Self-Service Submit receipts via Slack, ask about budgets, request payslips
Invoice Processing Auto-register supplier invoices from email, create customer invoices
Periodic Tasks Month-end closing, bank reconciliation, VAT return prep, payroll
Analytics Cost trends, outstanding amounts, budget vs actual

How It Works

Employees interact via Slack, Teams, web chat, or email. Messages are handled by the Conversation Agent, which manages multi-turn context, asks clarifying questions, and detects language (Norwegian/English). Once the user's intent is clear, the Conversation Agent sends a single classified task to the Accounting API (POST /solve).

Channels (Slack, Teams, Web, Email)
  → Conversation Agent (multi-turn, context, clarification)
    → Accounting API (stateless, one task per call, IDesign)
      → Tripletex / Fiken

The Accounting API reads uploaded documents (PDF invoices, receipt photos) and executes the correct accounting operations. A configurable rules engine ensures compliance with company policies and Norwegian accounting law (bokforingsloven). Every action is logged in an audit trail with undo capability.

Architecture

Built on the IDesign 4-layer architecture:

  • Clients — SolveController, SchedulerController, DashboardController
  • Managers — Task orchestration, agent loop, monitoring, audit
  • Engines — Rules, classification, calculations, compliance
  • Resource Access — Accounting providers (Tripletex, Fiken), LLM, notifications, database

→ View C4 diagrams

Built On Competition Experience

This product is based on learnings from NM i AI 2026 (Norway's National AI Championship), where we:

  • Built a working Tripletex AI agent scoring in the top 10% (33rd of 369 teams)
  • Studied 22 competing implementations from other teams
  • Identified the winning architecture: LLM agent with function-calling
  • Documented every Tripletex API quirk and scoring pattern

→ Technical details