We Built a Bookkeeping App You Can Just Talk To—Here’s What We Learned

Most bookkeeping software still feels like a maze of tabs, reports, and dropdowns. Even “simple” tools expect users to think like accountants. We wanted to flip that model: what if you could ask your bookkeeping software questions in plain English and get accurate answers?

This led us to build an AI-first bookkeeping app, where natural language querying serves as the primary interface.

The Architecture: SQL + Vectors

At the core, we use two complementary databases:

• A SQL database for structured financial data: transactions, invoices, categorization, and reports. This ensures that every number comes from a deterministic source.

• A vector database for general bookkeeping knowledge: definitions, how-to guides, accounting standards, and contextual explanations.

When a user asks a question, the AI doesn’t just guess. It interprets intent, decides whether to query structured financial data or fetch an explanation from the vector store, and then responds accordingly.

Example:

• “How much did I earn last quarter?” → SQL query, aggregated totals.

• “What’s a profit and loss statement?” → Vector retrieval, plus links to supporting resources.

• “What expenses can I cut?” → Hybrid: SQL-driven analysis + vector-driven suggestions.

This division provides us with accuracy in numbers and flexibility in explanations—without hallucinations sneaking in.

Building Conversational Bookkeeping

The challenge wasn’t just data architecture—it was UX. People don’t want to run “reports,” they want answers. Our AI translates everyday questions into actions and explanations:

• “How much did I spend on software subscriptions last month?”

• “Send an invoice to Client X for $2,500.” (coming soon)

• “Show me expenses trending up over the past six months.”

We have also added contextual help: if you ask, “What is cash vs. accrual accounting?” the AI provides a direct answer and links to more in-depth resources.

The Road Ahead: Agentic AI

The current version focuses on question answering and explanations. But the next version will go further: agentic AI capabilities.

Instead of just telling you the answer, the AI will be able to:

• Send an invoice.

• Generate a P&L report.

• Categorize new transactions.

• Run a reconciliation check.

This eliminates the constant tab-switching and menu navigation that slows down small business owners today. The ultimate goal: bookkeeping that feels less like “using software” and more like talking to an assistant who just gets it done.

Why This Matters

Small business owners who currently slog through QuickBooks input and manual categorization don’t just need prettier dashboards—they need frictionless workflows. By combining SQL for structured accuracy, vectors for contextual knowledge, and natural language for interaction, we aim to consolidate bookkeeping into a single, conversational interface.

And once agentic capabilities are live, most bookkeeping tasks—from invoicing to reporting—will move from “clicking around software” to “just asking and letting the system handle it.”