Choose, price and budget the AI
AI reads receipts, statements and screenshots and suggests categories. Settings → AI says which provider and model do each, what a call costs, how much the workspace may spend in a month and what it has spent.
- 1
Use included AI
On Personal or Pro, the workspace starts on the installation's own AI, with no key to bring, within the plan's monthly allowance, and says so the first time you open it. Settings → AI starts with Included AI: Stop using included AI turns it off, and Use included AI turns it back on. On Free, or for a model of your choosing, connect a provider of your own as below.
See also: Choose a plan and pay
- 2
Connect a provider
Under Settings → AI, Set up AI connects a provider in one step: OpenRouter unless Use another provider chooses another, its API key — Get an API key opens the provider's own page — and Connect and check, with a monthly AI budget of 10 USD. Add a provider sets one up by hand: a Name, a Provider key, a Protocol, a Base URL and its API key, and Discover models lists the models it offers. A model running on your own computer has ollama or lmstudio as its Provider key and a local address, such as http://localhost:11434/v1: it costs nothing, and is reached only from the desktop app.
- 3
Choose a model for each purpose
Select a model names, for a Purpose — Text, Vision or Transaction classifier — the Provider and the Model ID that does it, then Save selection. Then tick AI features enabled and Save: nothing is asked of a model until AI is on.
- 4
Set how transactions are classified
Transaction classifier suggests a category for a merchant the rules, memory and starter pack could not place, for review, never applied. Its Mode is Automatic, My selected classifier or Off; a Confidence floor keeps back what it is less sure of; Ask the chat model when the classifier is unsure gives it a second opinion. Then Save classifier settings. Its Privacy routing is said beside it: only providers that neither keep nor train on what is sent. Under Receipt lines, Suggest categories for receipt lines lets members ask the same classifier about a receipt's lines.
See also: Categories, rules and suggestions
- 5
Price each model
Add a price version names a Provider, a Model ID and a Currency, with the Input and Output price per 1M tokens, as the provider publishes them. A price takes effect when it is added and is never edited: a new price is a new version, and calls already costed keep theirs. A model without a price is flagged under Model selection — its calls are held at the reserve, not counted as spent.
- 6
Set a monthly budget
Add a budget takes a Currency, a Monthly limit and a Per-attempt reserve: what each call holds against the limit until its cost is known. When what was spent and what is held leave no room, new AI requests stop until the month resets. A local model costs nothing and never spends from it.
- 7
Read what it has cost
Usage lists each call: when it Started, its Model, its Status and its Cost — from the price, or reported by the provider. Above it is this month's total in the budget's currency, and whether every call is accounted for. Disable AI on the status card stops every AI feature; the connection, models and budget stay saved.