Spend by provider
Billed and estimated, never mixed.
- Anthropic$25.1K$23.9K billed · $1.1K estimated
- OpenAI$13.9K$13.2K billed · $723 estimated
- Google Gemini$6.2K$5.7K billed · $504 estimated
Billed Estimated · provider usageStriped: not billed yet
Free beta
Know what every AI dollar buys.
FP&A for AI spend, from provider bill to customer margin.
Connects to
Product
From invoice to answer.
Cost per customer
Every request tied to who it served.
| Customer | Cost | Share of revenue |
|---|---|---|
| Acme Corp | $13.1K | 8% rev |
| Globex | $9.3K | 11% rev |
| Initech | $5.9K | 5% rev |
| Umbrella | $3.8K | 14% rev |
| Unattributed | $2.9K | n/a |
Model comparison
Your token mix, priced on other models.
Mix: 2.1B input, 0.6B output tokens a month
claude-sonnet-5-5$10.2K
Current · $2 / $10 per 1Ma month
Source platform.claude.com
Read 2026-11-08
gemini-3.8-flash$3.8K
Google · $0.75 / $3.75 per 1Ma month
Source aistudio.google.com
Read 2026-11-08
List prices only. Quality is compared in evaluations.
Forecast with a range
p10 to p90, not a single guess.
Actual Forecast p50 p10 to p90
Budgets and alerts
Warned before the month ends.
- Month-end forecast
- 92%
- Alert at
- 90%
support-bot · November
Ask the analyst
Answers cite the numbers they used.
Why did spend rise last week?
Spend rose 14% week over week. Most of the increase came from the support-bot feature on claude-sonnet-5-5[1], mainly for Acme Corp[2].
- 1analytics.breakdown · feature · Nov 1 to 7
- 2customers.top · cost · Nov 1 to 7
Which customers cost most per request?What if we move search to gemini-3.8-flash?
Sample data Tiles show a sample workspace with synthetic figures.
Security
Read-only by design.
- Read-only keys and database roles.
- Credentials encrypted, never shown again.
- Every request checks workspace membership.
- No prompt or completion text by default.
- Sensitive columns start unselected.
- Audit log for connection and member changes.
FAQ
Questions, answered.
- What is AI Spendbook?
- AI Spendbook is financial planning and analysis (FP&A) software for AI spend. It shows what a company spends on AI providers, attributes the cost to customers, features and teams, compares models on the company's own usage, forecasts spend with ranges, and answers questions with citations to the underlying data.
- Which AI providers does AI Spendbook support?
- Anthropic and OpenAI connect through their admin usage and cost APIs. Google Gemini connects through the Cloud Billing export to BigQuery. Any other provider can be added with CSV or JSON imports, the Python SDK (pip install spendbook) or the HTTPS ingestion API.
- How does AI Spendbook reconcile provider billing with app telemetry?
- Provider billing sets the total for each day and provider. Request-level telemetry from your app or the SDK explains that total by customer, feature and user. The two are never added together. Billed spend that telemetry does not cover is shown as unattributed, and days the provider has not billed yet are labelled as estimates.
- What data does AI Spendbook read?
- Usage and cost reports from provider admin APIs, the Cloud Billing export for Gemini, the tables or collections you map in Supabase, Postgres or Firestore (only the columns you approve), Stripe invoices if you connect revenue, and events you send with the SDK or API. It only reads from these sources.
- Is AI Spendbook free?
- Yes. AI Spendbook is free during the beta, with every feature included and no card required. If paid plans are introduced later, users will be told before anything changes.
- Does AI Spendbook store prompts or completions?
- No, not by default. The SDK sends token counts, model, latency, status and attribution fields. Prompt and completion text is captured only if a workspace admin turns on content capture and the SDK is configured to send it.