The Ask AI feature expands from a read-only question-answering tool into a full action-capable assistant — able to write ledger entries, accept uploaded images, and more.
Beancount.io today announced a major expansion of its AI assistant.
When Ask AI launched, it answered questions. You could ask about your spending, query account balances, explore your ledger through natural language — and the AI would respond with information drawn from your books. It was a reader, not a writer. A powerful one, but bounded: if you wanted to change something, you were on your own.
That changes now.
The expanded assistant can write. Ask it to fix a misclassified entry, draft a new transaction from your description, or generate a batch of entries for a set of expenses — and it will produce the exact ledger text, show it to you, and wait for your approval before anything is committed. Nothing changes in your books without your say-so.
It can also read images. Upload a receipt, a bank statement screenshot, or a scanned invoice. The assistant reads what is in the image, connects it to the right accounts and categories based on your existing ledger patterns, and drafts the appropriate entry for your review.
This is the same assistant you already use. Same conversation interface. Same context that carries across your session. The difference is that it can now close the loop — from question to understanding to action — without you having to leave the chat.
"Ask AI was built to help you understand your finances. Now it can help you maintain them."
— the Beancount.io team
The expanded assistant is live at /agent.
Try It Now
Ask AI is live for all beancount.io users. Open your ledger and navigate to Ask AI — the URL follows the pattern:
https://beancount.io/ledger/{your-username}/{your-ledger}/agentThings to try:
- "What did I spend on dining last month?" — read-only query, instant answer
- "Add a transaction: $42 at Whole Foods on June 20, groceries" — the assistant drafts the entry and waits for your approval
- Upload a receipt photo — the assistant reads it and proposes the correct ledger entry
Every write is shown as a preview before anything is committed. Nothing changes in your books without your explicit confirmation.
Frequently Asked Questions
Q: What could Ask AI do before, and what can it do now?
Before: Ask AI could read your ledger and answer questions. Balances, spending trends, account history, transaction searches — anything that required reading your books but not changing them.
Now: it can also write. Propose a new entry, correct an existing one, or record a transaction from a description or an uploaded image. Every write is shown to you as a preview before it is committed. You approve or reject it; nothing is automatic.
Q: Where does my ledger and image data go?
Your books stay yours. The assistant reads only what it needs to answer your question or draft the entry you asked for, and nothing is written or committed without your explicit approval. Uploaded images are used to produce that draft — nothing more.
Q: What kinds of images can I upload?
Receipts, invoices, bank statement screenshots, or any document that describes a financial transaction. The assistant reads the document, identifies the relevant details, and drafts a Beancount entry. If anything looks wrong — a date, an amount, an account — you can correct it in the conversation before approving the write.
Q: Is there any risk of the AI editing my ledger by mistake?
No. Every proposed edit is shown to you as an explicit before/after preview. The assistant cannot write to your ledger without a confirmation step. This is not a setting you can accidentally turn off — it is how the approval flow is designed.
Q: Will my existing Ask AI conversations change?
No. Your conversation history and familiar interface carry over. The expanded assistant is the same Ask AI experience, now at /agent.
Q: What if the AI misreads a receipt or drafts the wrong entry?
Review it before approving. The draft is shown in full — account names, amounts, dates, narration — before anything is committed. If something is wrong, tell the assistant what to fix and it will revise the draft. Nothing is written until you say so.
Q: How does the approval flow for writes work?
When the assistant determines a write is appropriate, it produces a structured edit proposal: the file, the original text (if modifying), and the replacement text. This is shown to the user as a preview in the conversation. The user confirms or declines. Confirmation triggers the write and commit. Declination returns to conversation without any change. The assistant never writes without an explicit confirmation step in between.
Q: Does image upload require changes to how the assistant accesses ledger data?
No. Image input is a new input modality — the assistant can now accept an image alongside text. What it does with the image is the same kind of reasoning it already applies to text: extract relevant financial information, match it against ledger context, and propose an appropriate entry. The ledger access model is unchanged.
Q: What does this unlock on the roadmap?
Image upload for receipts is the first input modality beyond text. The same infrastructure supports any document type that carries financial information — invoices, bank statement PDFs, expense reports. The write capability is the foundation for features like assisted bank transaction reconciliation (via Plaid) and bulk entry generation from imported files. None of those require rearchitecting the assistant — they add new inputs or new action types to an existing loop.