ActivityLog collects usage from AI coding tools and assistants: sessions, tokens and cost, per person and per model. You see adoption and spend side by side.
Your sessions, what each one cost, and where the expensive ones were.
AI usage filtered to Michael Suyama: his sessions for the week with model, tokens, cost and duration, from Claude Code's own reporting. Tuesday's long session is the expensive one.
A single session: model, duration, cost and tokens at the top, then a waterfall of each prompt and model request over time, with the tokens and cost of each step. It shows where a long session spent its money. Prompt text is not shown.
AI cost per person, team and model, who is using the tools and who is not, and how spend is trending.
Headline numbers for the week (total cost, active users, sessions, tokens and error rate), then sessions over time and the most recent sessions with person, model, tokens and cost. The long, expensive session on Tuesday stands out in both the density chart and the session list.
One row per person for the quarter, largest spend first, built from the cost Claude Code reports with each request. Read it next to adoption: high spend can mean heavy, productive use, and people missing from the list are not using the tools at all.
Cost for the quarter split by person and by model, each block sized by what it cost. A team running everything on the largest model is a quick place to save money.
AI usage comes straight from the AI tools. Claude Code reports each session, model, token count and cost as people work, tied to the person who signed in.
Sessions, prompts sent, the model used, input and output tokens, cost per request, duration and which tools Claude used. Prompt text only if you choose to send it.
Claude Code has built-in OpenTelemetry reporting. Set a few environment variables on each machine, or in a shared settings file, with an ActivityLog token. It works for subscription seats (Pro, Max, Team) as well as API keys.
Model, tokens, cost and duration from any AI tool or app that reports in the OpenTelemetry generative-AI format.
Point the tool's OpenTelemetry exporter at ActivityLog with a token.
Each request, prompt and tool call is reported within seconds, with tokens, cost and model.
Claude Code includes the signed-in user's email on every record, so the whole team can share one token and still show up as separate people. If someone's Claude account uses a different email, it can be added as an alias.
Cost per day, person and model, sessions and their length, tokens by type and the tools Claude used most.
See who has adopted the tools and who has not, find the sessions that cost the most, and track spend over the quarter.
No. Prompt text is only collected if you turn it on, and it is never shown on any report. If it is collected, it is scanned for secrets and personal data, and only the category of a finding is kept, never the text that matched.
Yes. The data comes from Claude Code on each machine, not from Anthropic's billing, so Pro, Max and Team seats are covered as well as API keys.
Claude Code works out cost per request from its own price list. Credits, plan pricing and rounding are not in that figure, so treat it as a close guide to spend, not an accounting number.
ActivityLog is in beta. Connect the tools you already use and get this report for your organisation.
See pricing →Questions first? Talk to us → · All use cases
