Skip to main content

lim — Accounting as a Function

lim is a real-time accounting engine where AI agents — not humans — process business events into journal entries automatically. Bank transactions, invoices, and expenses flow in continuously. A judgment engine decides how to categorize each one. The system learns from human corrections. Within 6 months, 95% of transactions need zero human input.

How It Works

1

Events flow in continuously

Bank transactions via API, invoices via webhook, expenses via MCP tool calls from AI agents. Each event is normalized to a common format.
2

The Judgment Engine decides

Three steps, in order:
  • Rule Match — deterministic rules match patterns (e.g., “AWS monthly bill → Server Costs”). Zero AI cost.
  • History Match — SQL lookup of past similar transactions.
  • AI Inference — only called when rules and history can’t resolve.
  • Escalate — low-confidence items go to a human for confirmation.
3

The system learns

When a human confirms an AI suggestion, it becomes a rule. After 3 confirmations, the rule auto-posts. AI costs approach zero over time.
4

Financial data is always live

Trial balance, P&L, cash runway — all computed from the journal in real-time. Every change is audit-logged.

Architecture

┌─────────────────────────────────────────────────────┐
│  Layer 1: INGESTION                                  │
│  Bank APIs │ Webhooks │ MCP Tools │ Manual Entry     │
├─────────────────────────────────────────────────────┤
│  Layer 2: JUDGMENT                                   │
│  Rules (95%) → History (3%) → AI (1%) → Escalate    │
├─────────────────────────────────────────────────────┤
│  Layer 3: LEDGER (Audit-Logged)                      │
│  Journal Entries │ Debit=Credit │ Full Change History │
├─────────────────────────────────────────────────────┤
│  Layer 4: QUERY                                      │
│  Trial Balance │ P&L │ Cash Runway │ Scenario Diff   │
├─────────────────────────────────────────────────────┤
│  Layer 5: OUTPUT                                     │
│  MCP Server │ REST API │ Webhooks │ Dashboard        │
└─────────────────────────────────────────────────────┘

Key Features

Natural Language Input

Describe transactions in plain language. lim handles the accounting.

AI Agent Protocol

Native MCP server. Any AI agent that speaks MCP can do accounting through lim.

Learning Engine

Rules → History → AI → Escalate. The system learns your patterns over time.

Full Financial Reports

Trial Balance, P&L, Balance Sheet, Cash Flow, Shareholders’ Equity — all real-time.

Japanese Compliance

Consumption tax, e-Tax XML export, Invoice System (インボイス制度), Electronic Bookkeeping Act (電帳法).

Scenario Analysis

“What if we hire 2 engineers?” — virtual journal entries show the impact on runway instantly.

Next Steps