Frequently asked questions
Short answers grounded in how Fleet actually works. For the install path see Install & Connect. For the graph model see Workflows.
What is Fleet?
A local Go binary plus a hosted control plane. Fleet-launched Linux agents run in a jail. They cannot use your keys or open the network until you allow a host. It also runs saved coding-agent workflows: typed steps, triggers, approvals, and provenance. Agents run through Claude Code or OpenCode. Grok Bot, Cursor Cloud, and Codex cloud VMs run on their vendors.
How do I prove agents cannot escape?
Run fleet doctor, then go test -count=1 -run TestAcceptance_CompromisedAgentHasNoAlternateEgressPath ./internal/sandbox/. A passing run prints TCP_DENIED, SECRET_DENIED, and the rest of the transcript. Write-up: https://fleetctl.ai/security/#containment.
Which coding agents does Fleet use?
Agent steps run through Claude Code or OpenCode, including models those harnesses support. Fleet starts the session, routes work, and records the run. It does not replace the harness you already authenticated.
Do I need Docker?
No. Install the official binary. You need tmux, git, and Claude Code or OpenCode on the machine that runs Fleet. The GitHub CLI is required for coding workflows that open and review pull requests.
Does Fleet send my source code to the control plane?
No. Source code stays on your machine and goes only to your model backend and GitHub. When you connect Fleet to the dashboard, the watcher reports operational metadata — agent status, run counts, run duration, fabric events — plus one billing fact per durably authorized hosted root-workflow start.
Is Fleet silent until I register?
An unregistered instance does not sync metering to the dashboard. The CLI still sends opt-out usage analytics (command names, version, OS — never code, paths, args, or repo names) unless you disable it with fleet config set telemetry off, FLEET_TELEMETRY=0, or DO_NOT_TRACK=1.
What does the free tier include?
500 hosted root-workflow starts. No card. No calendar cutoff. A start is recorded when the first step of a root workflow is durably authorized, even if the workflow later fails. Steps, retries, reviews, approvals, and same-run resumes do not multiply usage.
How does paid pricing work?
Business is $299/org/month or $2,990/year (unlimited users and repositories, 25,000 monitored starts). Enterprise is a custom annual contract. Paid allowances are monitored with 50/80/100% alerts and do not trigger automatic overage charges or interruption. Model-provider fees are separate.
What is a workflow versus a run?
A workflow is the saved graph: trigger, typed steps, dependency edges, outcome routes, and governance. A run is one execution of that graph. History keeps status, artifacts, decisions, retries, and provenance.
How do I start?
Install the binary, run fleet up in the repository Fleet will work in, sign in at app.fleetctl.ai, save a workflow (Pull request code review, Ship a feature, or Weekly brief), and press Run.
Where does a run execute?
On your machine. Workflows are authored in the dashboard builder and pulled by the local worker started with fleet up. Source code stays on that machine and goes only to your model backend and GitHub.
Do I need the Fleet Reviewer GitHub App?
Yes for automated pull-request review. Source-control actions use the GitHub CLI; pull-request workflow triggers use HMAC-verified GitHub App webhooks. Installing the Fleet Reviewer App is required for automated PR review.
Is there a marketplace of workflow templates?
No. The builder ships a small gallery for cold-start — Pull request code review, Ship a feature, and Weekly brief — not a marketplace of prebuilt workflows.