CloudByte AI shows how your team uses Claude Code: every session, prompt, tool call, model and token, per developer and project, next to what it costs and the commits it produced. One install line per laptop, live in about 10 minutes.
What can you see about your team's Claude Code usage?
Every Claude Code session on every developer machine, with its prompts, tool calls, model and tokens, added up per developer, project and team. Open any session to read it from the first prompt to the commit it ended in.
Prompts and responses in order, with developer, project, repository and branch
Tool calls recorded as observations: Bugfix, Feature, Refactor, Change, Discovery and Decision
Input, output and cache tokens, and the model behind every response
An AI summary of each session, and the commits it led to
Each token is priced at the list price of the model that used it, then added up per session, developer, project and model. You see where spend goes, and which sessions and projects drive it, before the invoice arrives.
Input, output, cache write and cache read tokens kept apart
Spend grouped by developer, project or model, with daily trends
Cache hit rates, so sessions that keep re-reading context stand out
On the Enterprise plan, BYOK keeps spend on your own Anthropic account
AI Insights weighs the engineering time saved against what Claude Code cost, for each team and project, and shows the adoption behind it: who uses it every day, how deep sessions go and how many end in a commit.
ROI per AI dollar
An hours-saved estimate set against token spend, for each team and project.
Adoption you can trust
An adoption heatmap and active days per developer, with broken installs kept apart from real non-use.
Sessions that ship
Session depth and commit conversion by project, so you see which work AI actually moves forward.
ROI and spend analytics are included from the Starter plan.
CloudByte AI checks whether each project has a CLAUDE.md and whether Claude Code sessions follow it, and records which shared skills ran and when developers overrode them. A spike in overrides shows which skill needs an update.
CLAUDE.md compliance per project and per developer, on the Team plan
Usage counts and override rates for every shared skill
1,000+ approved skills, matched to each project by Skill Harness
Project skills pushed to every developer and kept in sync
Each agent sends a heartbeat every 3 minutes, so a broken install never looks like a developer who stopped using Claude Code. Low adoption and a silent failure are easy to tell apart.
Online, Stale and Offline status for every machine
Version drift and staged rollouts across the fleet
Alerts by email or Slack within 10 minutes of an agent error
Hook crashes are isolated and never crash the Claude Code session
Install the agent with one line per laptop. It reads Claude Code's local session data without changing it, syncs every 3 minutes, and your dashboard is live in about 10 minutes.
01
Install once per laptop
One command on macOS, Linux or Windows. The installer finishes in 60 to 90 seconds and needs no change to any repository.
02
Keep using Claude Code
Developers use Claude Code exactly as before. The agent runs in the background and never sits between them and the model.
03
See the data in minutes
Prompts appear within 3 minutes of the first session and commits within 10, per developer, project and team.
What the agent sends
Prompts and tool observations
Session details and summaries
Token usage per model
Commit metadata and diff stats (diffs kept 5 days)
Claude Code's built-in usage views vs CloudByte AI team analytics
Anthropic gives organizations real usage reporting for Claude Code: adoption, accepted lines and spend, with contribution metrics on Team and Enterprise plans. CloudByte AI adds session-level detail, commits, project context and agent health around it, and puts Cursor in the same view.
How Anthropic's built-in Claude Code usage views and CloudByte AI compare
Topic
Anthropic's built-in views
CloudByte AI
Where you see it
/usage in the terminal for your own session on one machine, plus a Claude Code analytics dashboard in claude.ai (Team and Enterprise) or the Claude Console (API)
One web dashboard for the whole team, fed by an agent on every developer machine
Adoption
Daily active users and sessions, with a contributor leaderboard on Team and Enterprise
Active developers, sessions and prompts per developer, project and team, with broken installs flagged
Spend
Spend per member in the Claude Console; on Team and Enterprise, a spend report for usage beyond the seat allowance
Input, output and cache tokens with an estimated cost per developer, project and model
Code output
Lines accepted and suggestion accept rate, plus PRs and lines shipped with Claude Code through the GitHub app on Team and Enterprise
Local and remote commits with an AI summary, linked to the sessions and tickets behind them
Project context
On paid plans, /usage shows how much of your own recent usage went to skills, subagents and MCP servers
CLAUDE.md compliance per project on the Team plan, and usage and overrides for every shared skill
Other AI tools
Claude usage only
Claude Code and Cursor in one view
Data access
CSV export, OpenTelemetry export, and the Claude Code Analytics API (Console) or Enterprise Analytics API
Every metric over a REST API with an OpenAPI 3 spec
Where you see it
Anthropic's built-in views
/usage in the terminal for your own session on one machine, plus a Claude Code analytics dashboard in claude.ai (Team and Enterprise) or the Claude Console (API)
CloudByte AI
One web dashboard for the whole team, fed by an agent on every developer machine
Adoption
Anthropic's built-in views
Daily active users and sessions, with a contributor leaderboard on Team and Enterprise
CloudByte AI
Active developers, sessions and prompts per developer, project and team, with broken installs flagged
Spend
Anthropic's built-in views
Spend per member in the Claude Console; on Team and Enterprise, a spend report for usage beyond the seat allowance
CloudByte AI
Input, output and cache tokens with an estimated cost per developer, project and model
Code output
Anthropic's built-in views
Lines accepted and suggestion accept rate, plus PRs and lines shipped with Claude Code through the GitHub app on Team and Enterprise
CloudByte AI
Local and remote commits with an AI summary, linked to the sessions and tickets behind them
Project context
Anthropic's built-in views
On paid plans, /usage shows how much of your own recent usage went to skills, subagents and MCP servers
CloudByte AI
CLAUDE.md compliance per project on the Team plan, and usage and overrides for every shared skill
Other AI tools
Anthropic's built-in views
Claude usage only
CloudByte AI
Claude Code and Cursor in one view
Data access
Anthropic's built-in views
CSV export, OpenTelemetry export, and the Claude Code Analytics API (Console) or Enterprise Analytics API
CloudByte AI
Every metric over a REST API with an OpenAPI 3 spec
Anthropic's side is based on its public Claude Code documentation, checked in October 2026. Features change often, so check Anthropic's documentation for the latest. Product names are trademarks of their respective owners.
FAQ
Claude Code analytics questions, answered
How tracking works, what it costs, what it captures and what it never touches.
Install the CloudByte AI agent once per developer laptop with a single command. It reads Claude Code's local session data without modifying it and syncs prompts, tool calls, tokens, models and commits every 3 minutes, so one dashboard shows usage per developer, project and team, alongside Anthropic's own analytics.
Yes. Every session records its input, output, cache write and cache read tokens and the model behind each response. The AI Cost Center adds them up per developer, project or model, with an estimated cost at list prices. Token spend and ROI analytics are included from the Starter plan.
Claude Code itself can use an Anthropic API key instead of a Claude subscription. With CloudByte AI BYOK on the Enterprise plan, requests run through your own Anthropic account, so spend lands on your Anthropic bill and CloudByte never proxies billing. The key is encrypted at rest with AES-256-GCM.
No. The agent only reads session files that already exist and sends small batches of metadata on a short interval. It never sits between the developer and Claude Code, so prompts and responses are not delayed, and a crashed hook is isolated, so it never takes a session down with it.
Prompts and tool observations, session details and summaries, token usage per model, commit metadata with diff stats, and agent heartbeats. Commit diffs used for summaries are kept for only 5 days. Source code files, keystrokes, environment variables, screen recordings and browser activity are never collected.
