Every prompt and response is scanned on the developer machine for personal data, credentials and other sensitive information. You decide what counts as sensitive and what happens when it is found.
CloudByte AI reviews each project and suggests guardrails that fit it. Adjust them, save them as your team policy and push them to every developer machine in one step.
The agent discovers the AI coding tools, assistants and extensions running on each machine, approved or not, so nothing sits outside your policy.
Plugins and MCP servers can give an AI assistant access to databases, file systems and internal tools. CloudByte AI detects every one that is installed and flags the ones you have not approved.
A small change to Claude Code, Cursor or Copilot settings can quietly widen what an assistant is allowed to do. CloudByte AI watches these settings and alerts you when they change outside your policy.
Prompt alerts, guardrail hits, shadow AI, plugins and config changes all report to one dashboard you can filter by team, project, developer and severity.
5
Detection types in one view
3 min
Agent sync interval
0
Workflow changes for developers
How prompt scanning, guardrails and detection work on your team.
On the developer machine, before the prompt is sent to the model. A prompt that breaks a blocking rule never leaves the laptop, and admins are alerted in real time.
Yes. PII, credentials, API keys, tokens and private keys are covered out of the box, and you can add your own rules for things like customer IDs or internal hostnames. Each rule can block or flag, with its own severity.
CloudByte AI looks at each project and its stack and suggests guardrails that fit it, such as which folders the assistant may edit. You review and adjust them, save them as a team policy, and push it to every team member in one step.
Any AI coding tool, assistant or extension running on a developer machine that your team has not approved. The agent builds an inventory per machine and flags new tools as soon as they appear.
Every plugin, extension and MCP server installed for the AI tools on each machine, along with what it can reach, such as a database or the file system. Anything not on your allowlist is flagged.
Claude Code, Cursor and GitHub Copilot settings and permission files, such as the allowed tools list and MCP configuration. You see exactly what changed, line by line, and admins are alerted when a change breaks your policy.
Not in day-to-day work. It runs inside the same lightweight agent that tracks AI activity, with no change to how developers code. They only notice it when a prompt or change breaks a rule.
AI Security and shadow AI detection are included in the Team plan and in Enterprise.
Install the agent once per laptop and see prompt alerts, plugins and config changes on one dashboard within minutes.
