Security

Your agents, your machines, your account. Fleetly just coordinates.

Agents with access to your code and your servers call for care. Here’s what lives where, what goes through the hub and how each part is protected, no fine print.

Where everything lives

Stays on your machine
  • Your project code and files
  • The Claude Code, Codex, Antigravity and GitHub Copilot sign-ins
  • The running agents, on your account
  • Your git and server keys
  • Routines and what each run did
Goes through the hub
  • Session state (who’s working, who’s waiting on you)
  • The conversations you open in the dashboard
  • The files you open or save in the Files tab (secret files such as .env don’t open)
  • Approval requests and your answers
  • The preview pages you open
  • The morning brief, put together when it’s sent (only the days and time are stored)
The hub stores only a hash
  • Each machine’s token
  • Recovery codes
  • Browser sessions
  • Invite links

Everything that goes through the hub travels over an encrypted connection. Your Claude Code, Codex, Antigravity and GitHub Copilot credentials are never read: the collector only checks that they exist.

Signing in

No Fleetly password

Sign in with a passkey (Face ID, Touch ID, Windows Hello or a security key) or your Google, Microsoft or Apple account. There’s no Fleetly password to leak.

Phishing sites don’t work

The passkey is bound to Fleetly’s address. On a page imitating the sign-in, it simply won’t open.

Rate-limited attempts

Repeated failures pause attempts from that network for a few minutes, without affecting anyone else.

Single-use recovery

8 codes shown once, stored only as hashes. Each one works once.

Connecting machines

Outbound connections only

The collector calls the hub, as your user (no root or administrator). No open ports, no public IP, works behind firewalls and NAT.

One token per machine

Revoke one machine and only that one is cut off. On the machine, the token is readable only by your user.

Verified collector

The collector updates itself and checks the hash of every file before running it.

You choose the folders

On a laptop with personal stuff, the machine shows the building only your work folders. The machine decides: the hub can’t widen it.

Controlling what happens

You approve what runs

Claude Code permission requests are still yours. Fleetly brings the request to you.

Roles

Owner, operator or viewer, per building. Viewers can’t send messages or approve.

Everything audited

Sign-ins, messages, approvals, commits, pushes and settings changes, with the person, address and time.

Instant lockout

Locking an account kills its sessions on the spot, including open previews.

Company governance (Enterprise)

Everything about governance →

Rules before every action

The building’s policy applies on the machine itself, before every command, edit or MCP call, in auto mode too and even with nobody on the dashboard.

Blocked commands

The owner lists what never runs (like git push --force), and the machine refuses right away. Also on Team.

Two-person approval

Sensitive commands, like a deploy, and pushes to protected branches only run once two different people approve.

Where and when agents act

Blocked folders or only allowed ones, and working hours: outside them, the agent only reads.

Network, models and MCP

The agent only reaches approved domains, uses the allowed models and the MCP servers on the list. Below the minimum Claude Code or Codex version, it doesn’t act.

Secrets masked at the source

Keys, tokens and passwords become ••• on the machine itself, before any conversation or output leaves for the hub.

Simulate before it applies

Before saving, see how many times the policy would have blocked or asked for approval over the last week.

SSO and SCIM

Company sign-in (Entra ID, Okta, Google Workspace) with a verified domain. The directory adds and removes people, and removing someone ends their sessions immediately.

Agent trail

Every agent action on the machines, with who authorized it and whether it was through the dashboard or on the machine.

Audit proof outside the database

Records are chained, and every day a signed copy goes to locked storage outside the database that nobody can delete or overwrite during retention. The signed attestation shows whether anything was touched, even by someone with database access.

Protection people can’t turn off

IT installs Fleetly’s hooks in Claude Code’s managed settings (through MDM), and whoever uses the machine can’t remove them. If a machine’s protection drops, the building gets an alert, and the policy can make it read-only.

Straight into your SIEM

The audit log and the agent trail go to Splunk, Datadog or a signed webhook every 30 seconds. Or pull them through the API with a read-only token.

Documents for your review

Data processing agreement (DPA), subprocessor list, backup and recovery, incident response and a completed security questionnaire: ask at admin@fleetly.tools.

Isolated preview

Separate from the dashboard

The app under test was just written by an agent, so it runs on its own address. It can’t reach your session or the dashboard APIs.

Single-use ticket

Opening the preview creates a ticket valid for 60 seconds that turns into access to that address only.

Only the port you opened

The collector only forwards to the open preview’s port. Never to another service on the machine, not even if the hub asks.

Closes on its own

After 30 minutes idle, or 8 hours after it was opened. Opening one requires the operator role and is recorded in the audit log.

Routines without surprises

They run on your machine

Routines are stored on the project’s machine and run there, even with the dashboard closed. The hub doesn’t store what they do.

Push and PR are always yours

A routine never pushes or opens a pull request, even with accept automatically on.

The policy still applies

Blocks and two-person approval apply to routines too. Without accept automatically, each request waits for you up to 15 minutes and is then refused.

All in the audit log

Creating, changing, running and deleting is for building operators and goes to the audit log. Each run opens a session tagged rotina.

What about the command center?

The agent that sees the fleet runs on one of your machines, not on the hub. It queries Fleetly with temporary access of yours, valid for 2 hours, with the same scope and role as you.

Found a security issue?

Email admin@fleetly.tools with the steps to reproduce it.

Ready to connect your first machine?