Trustwelt AI Security · AI agents on the computers

Know which agents are on your computers, and what they can reach.

An AI agent runs as the person. Whatever the person can read, the agent can read: the SSH key without a passphrase, the AWS credentials file, the API key in a project's .env, the token in an MCP server's settings. Trustwelt Desktop looks once a day, in the places each tool documents, and never reads a value out.

A module of Trustwelt AI Security, included with it. Runs on: Managed computers with Trustwelt Desktop and its browser extension: the Chromium family and Firefox on Linux, Firefox on Windows; the AI agent gateway on managed Linux computers; other programs through the network filter, sites only.

Tenant Admin › AI services › Agents: five computers report, three with an AI agent, two with MCP servers, three with secrets in files; a table of computers with their agents, MCP servers, secret files and when each looked.

What it does

AI agents on the computers, in plain words.

  • Agents

    Found by name and version

    Command-line agents, editor extensions, IDE plugins and desktop apps: Claude Code, Codex CLI, GitHub Copilot, Cursor, Windsurf, Gemini CLI, Cline, Roo Code, Aider, JetBrains AI and Junie, Claude Desktop, ChatGPT.

  • MCP servers

    What the agents' settings start or call

    From each agent's own configuration files: the server's name, whether it runs on the computer or is reached over the network, the program's name or the host; never an argument, a path or a value. Keys written into an entry are counted.

  • Secrets in files

    Thirteen kinds, counted

    Agents' own sign-ins, AWS, Google Cloud and Azure credentials, kubeconfigs, .env files in projects, SSH keys without a passphrase, git and package-registry tokens, container-registry passwords. Kinds, places and counts only; buffers wiped.

  • Two views

    Counts for the organisation, findings for the person

    Tenant Admin sees the agents, the MCP servers and how many secrets of each kind, per computer. The person sees each finding in their Portal, where it is, and what to do about it. A person's own computer reports only when they choose.

How it works

From the console to the computer.

  1. Look once a day

    Ninety seconds after Trustwelt Desktop starts, then daily, and on "Look again": known places per operating system, no walk of the disk; project folders to a fixed depth and count, with a time limit.

  2. Report counts

    A managed computer whose organisation holds AI Security sends agents, MCP servers and secret counts; every field is checked again on the server.

  3. Read it in Tenant Admin

    AI services › Agents: computers with an agent, with MCP servers, with secrets in files; filter by agent; CSV and PDF exports, each export audited.

  4. Fix it from the Portal

    My agents lists each secret with its file and the step that removes it: move it into Trustwelt Desktop as an API key, SSH key, cloud role or database login, then delete the line or the file.

In Tenant Admin

What your administrators see.

Tenant Admin › AI services › Agents: five computers report, three with an AI agent, two with MCP servers, three with secrets in files; a table of computers with their agents, MCP servers, secret files and when each looked.
Agents in Tenant Admin: counts per computer, never a path or a value.

What is recorded, and what never is

Your organisation sees this. Nothing else.

  • Per computerThe agents installed with their kind and version, the MCP servers with name, transport and program or host, and the number of secret files and items by kind.
  • Exports and viewsWho exported the agent inventory and when, in the audit log.

Never:

  • A file's path, a key's name or a secret's value sent to the organisation: those stay with the person, in their Portal.
  • A person's own, unmanaged computer, unless they chose "Show in my Portal"; withdrawing it deletes the server's copy.
  • Any value read out of a file: the look reports kinds and counts, and wipes what it read.

Specification

Where it runs and how it is sold.

Runs on
Managed computers with Trustwelt Desktop; the places looked at are those each tool documents on Linux, macOS and Windows.
Agents known
Claude Code, Codex CLI, GitHub Copilot, Cursor, Windsurf, Gemini CLI, Cline, Roo Code, Aider, JetBrains AI and Junie, Claude Desktop, ChatGPT.
Limits
Project folders four levels under the home and two inside a project; at most 300 projects, 500 .env files, 256 KiB per file, 20 seconds.
Sold as
Part of Trustwelt AI Security.

Questions

Asked before buying.

Does Trustwelt read my .env file?
It reads the variable names and counts the lines that look like keys, on the computer, and wipes the buffer. The organisation receives the count; you receive the names in your Portal. No value is ever read out or sent.
What happens to the secrets it finds?
The Portal shows the step for each: move the secret into Trustwelt Desktop as an API key, SSH key, cloud role or database login, so agents use it without seeing it, then delete the line or the file and rotate what was shared.

More in Trustwelt AI Security

The other modules, and what works with this one.

Start with Trustwelt AI Security.

AI agents on the computers comes with Trustwelt AI Security, sold on its own per person per month. An organisation account is free.