Know what your AI agents cost.
See the work behind the bill.

No. It is comparison math based on published API rates. aibill keeps that separate from detected subscription context and optional provider-reported cost; final invoices may still include credits, discounts, tax, or adjustments.

Your transcripts, prompts, file names, and dollar amounts never leave your machine, and no account is needed. Two disclosed exceptions: the CLI counts which commands run — anonymous, never your data or content — after a printed first-run notice, and aibill telemetry off or DO_NOT_TRACK ends it; the optional launch-email ask sends exactly the one payload the CLI prints, only after you type y. Provider credentials are used only when you explicitly connect an official billing API, and MCP results go only to the AI client you invoke under that client’s data policy. aibill never sits in the inference path and never stores, prints, or proxies provider credentials.

The local CLI and explicit MCP integration are in public beta. Glance can be built from source for testing; a signed Mac download and the shared Workspace are not launched yet.

Locally, yes. aibill reads the limit windows your agents already report and shows runway — how much of the window is left and when it resets — in the CLI, in Glance, and in the Claude Code statusline. Detection is read-only from local state; where a source doesn't expose a limit, the gap stays visible instead of being guessed.

Not from spend evidence alone. The beta establishes cost, activity, attribution, and coverage, and now records locally confirmed ownership and accepted GitHub outcomes. Defensible ROI requires reconciled cost, an accepted outcome, and independently evidenced business value; the company-accountability layer that reconciles those at team scale is next.