SideNote Pro

Local usage and cost tracking for your AI API spending

Metered API billing is only uncomfortable when it is invisible. SideNote Pro keeps a local ledger of what you used, by provider and model, so you can see the shape of your spending without waiting for an invoice.

What gets recorded

Token counts
Local estimates, replaced by provider-reported figures when the endpoint returns usage metadata.
Provenance
Which provider profile and model a request went to, recorded immutably against the response.
Pricing snapshots
The price used for an estimate is recorded with it, so historical entries do not silently change when a price does.
A daily ledger
Bounded daily aggregates by provider, model and request category.
Local cost estimates
derived from the above.

Provider-reported usage takes precedence over local estimates whenever it is available. SideNote Pro recognises the common prompt, input, completion, output and total token fields in final stream events and response metadata. Endpoints that omit usage entirely remain fully supported; you simply get estimates rather than reported figures.

Settings › Usage & cost

Three usage cards, Today, Last 7 days and Last 30 days, each showing a token count and an estimated dollar figure.
Three windows onto the same ledger. Where a model has no price recorded, the card says so rather than quietly counting that usage as nothing.

Settings › Usage & cost

Recent provider and model totals, listing each provider and model with its token count and estimated cost, including one line reading cost not available.
Provenance is per provider and model, recorded against the response when it arrives, so a later change of default profile cannot reattribute work you already did.

Where prices come from

SideNote Pro ships no provider prices, not even for the default OpenAI model. Provider rates can change, so SideNote Pro uses prices you enter for local estimates instead of figures built into the application.

So set the price yourself. Each provider profile and model can carry an input and an output price in USD per million tokens, entered in Settings, AI providers. Prices you configure are labelled as user-configured, and historical responses keep the price snapshot they were recorded with. The GPT-6 Luna setup works through both figures for a current low-cost OpenAI model, along with what a typical request costs at those rates.

Settings › AI providers › Model pricing

The Model pricing block: input and output price boxes in USD per million tokens, above the Use these prices, Clear pricing and Reset counters for this model buttons.
Where the numbers come from: two boxes per model, in USD per million tokens, next to the sentence that explains why correcting one later does not rewrite the estimates already recorded.

Get the numbers from your provider's own pricing page at the moment you configure them. Model prices change, and a figure copied from a blog post six months ago will quietly make every estimate wrong.

The optional daily limit

You can set a daily estimate limit in USD. When the day's estimated spend would exceed it, SideNote Pro stops before sending, and it can stop an in-flight stream that crosses the line mid-response.

This is a guard rail against a runaway afternoon, not a billing control. It works from local estimates, so it cannot know about usage from other applications sharing the same API key, and it cannot reconcile against your provider's meter. Treat it as a seatbelt, and set a hard spending cap in your provider's own dashboard as the real limit.

Settings › Usage & cost

The Enable local daily usage limit switch turned off, the disabled Daily estimate limit box beneath it, and captions explaining that the limit is a local estimate and that the ledger stores only counts, dates, provider and model provenance and estimates.
The limit box stays disabled until the switch above it is on. The note under it gives the same warning: the limit works from local estimates.

Estimates, and what the ledger stores

SideNote Pro does not query a hosted provider's billing system. It sends the key you configured to authenticate requests, but has no access to the provider's account or charges from the usage page. A local endpoint may use no key at all.

Everything here is an estimate. For a hosted provider, its invoice is the authority. Estimates help you compare configurations; they do not reconcile a bill to the cent. A local model has no API charge, though its token counts still appear here.

The ledger stores counts, dates, provider and model provenance, request categories and cost aggregates. It does not store prompts, responses, selected text, filenames, document contents, screenshots, image data or API keys. Usage never leaves your PC and is never sent to BediniLabs.

What to do with the numbers

Use the ledger to compare a week of requests by provider and model. Look at which combinations dominate, and ask whether the expensive ones were doing work that needed them.

If a heavy model is handling quick rewrites out of habit, reducing your AI API costs shows ways to spend less, and choosing the right model helps match models to the work.

Keep reading