Through one lightweight agent of about 3 KB on each developer laptop. It reads the local session data of Claude Code and Cursor without modifying it, captures commits from the repositories you point it at, and syncs to your organization every 3 minutes. It never sits between a developer and their AI tool, so prompts and responses are not delayed.
The rest connects on the server. GitHub adds remote commits and pull requests, Jira and Linear tie sessions to tickets, Google Workspace SSO controls who signs in, and agent health alerts reach your team by email or Slack. On the Enterprise plan you can also bring your own model provider keys, so AI spend stays on your own account.
Each integration below says what it adds and, where it matters, which plan includes it.
Session capture from the AI coding tools on each developer machine.
Sessions, prompts, tool calls, models and tokens from Claude Code, per developer and project, with cost, CLAUDE.md compliance and agent health.
See the Claude Code integrationSessions, prompts, tool calls, models and tokens from Cursor, with cost and adoption per developer and Claude Code in the same view.
See the Cursor integrationLink AI sessions to the commits and tickets they produced.
Local commits from any Git repository, captured on the developer machine; remote commit and pull request sync for GitHub.
See Git ActivitySessions matched to tickets from branch names, commit messages and the session itself, so you see which ticket each developer is working on and how much AI work went into it.
See ticket trackingControl who signs in, and keep AI spend on your own accounts.
Google Workspace SSO restricted to your domain and magic-link sign-in on the Team plan and up. SAML, Okta and Azure AD on Enterprise, with four roles enforced on the server.
See identity and accessBring your own Anthropic, OpenAI, Google Gemini or AWS Bedrock key, encrypted at rest with AES-256-GCM. With your own Anthropic key, spend lands on your Anthropic bill and CloudByte never proxies billing.
See BYOKKnow when something breaks, see what is installed, and take the data with you.
See which AI coding tools and extensions are installed on developer machines, and flag the ones you have not approved.
See AI tool discoveryAgent health alerts by email or Slack when a machine goes Stale or Offline, and within 10 minutes when an agent reports an error.
See Agent HealthEvery dashboard feature is also a REST endpoint with an OpenAPI 3 spec and Bearer-token auth, so metrics flow into your own reporting tools.
See the API and audit trailWhich tools CloudByte AI works with, what changes for developers and how long setup takes.
Yes. The CloudByte AI agent reads the local session data of Claude Code and Cursor on each developer machine, without modifying it, and syncs prompts, tool calls, models, tokens and commits to one dashboard every 3 minutes. Both tools show the same metrics, so a developer who switches between them still appears as one person.
No. The agent is about 3 KB, installs with one line per laptop on macOS, Linux or Windows, and runs in the background. It only reads files that already exist and never sits between a developer and their AI tool, so prompts and responses are not delayed and nobody changes a single command.
About 10 minutes from sign-up. The installer finishes in 60 to 90 seconds per developer, an initial sync sends any backlog, and prompts appear on the dashboard within 3 minutes of the next AI session. Commits follow within 10 minutes, with nothing to configure in your repositories.
Install the agent once per laptop, connect GitHub, and see AI sessions, spend and commits on one dashboard.
