Your fleet, live.
A real-time, self-hosted web UI for Docker, Podman, remote hosts, Docker Swarm and Kubernetes, in one small binary.
curl -fsSL https://fleetops.run/install.sh | sh
Checks your machine, then starts FleetOps on http://localhost:9000. Apache-2.0. Windows and other ways · Pricing
Replay: one container failure, fixed in FleetOps
14:01:58. All 12 containers are running on prod-eu. Press play, or press j to step.
14:02:07. Docker reports oom, then die with exit code 137. The strip moves to Needs attention and its holder flashes. Nothing polled and nobody refreshed. Live by default
14:02:09. What needs you is a keyboard list. j and k move from strip to strip, and Enter opens the logs. Keyboard and ⌘K
14:02:11. Memory climbs to the 512 MiB limit and the log stops. The viewer keeps 200k lines, searches by text or regex, and resumes after a restart without repeating a line. The log viewer
14:02:31. Raise the limit and recreate the container. If the new one fails to start, FleetOps puts the old one back. Recreate with rollback
14:02:36. Docker reports start and the holder turns green. The audit log keeps who recreated it, and when. Roles and audit
Manages
- Docker and PodmanLocal socket or Podman's Docker API
- Remote hostsDial-out agent, no open port
- Docker SwarmAn agent on every node
- KubernetesKubeconfig or in-cluster agent
- HelmInstall, upgrade, roll back
Thirty seconds, five things it did right.
Each moment in the replay is a feature you would use on a bad afternoon.
Live by default
Lists and detail pages follow Docker's own events over one websocket. Only the rows on screen subscribe to stats. When the connection drops, the page says so and how old its data is; it never passes stale state off as current.
- Updates
- Docker and Swarm events, Kubernetes watches
- Polling
- None for lists
- Disconnected
- Said plainly, with the data's age
Hands stay on the keys
The attention bay is a list you walk, not a table you scan. The command palette finds any container, image or Kubernetes object and runs an action on it.
- j k
- Next and previous strip
- ↵
- Open its logs
- ⌘ K
- Jump to anything, act on it
- /
- Filter the list
A log viewer that keeps up
Virtualized to the newest 200,000 lines, with follow mode that pauses when you scroll up. A reconnect or a restart picks up where the log left off, without repeating lines.
- Search
- Text or regex, next and previous
- Streams
stdoutandstderrtoggles- Colour
- ANSI, as the app wrote it
- Export
- Lines shown, or the full log
Honest when it matters
Recreate a container with changes and, if the new one fails to start, FleetOps rolls back to the old one. The terminal never silently starts a new shell after a dropped connection, and Docker's error messages are shown exactly as Docker wrote them.
- Recreate
- Rolls back on a failed start
- Duplicate
- Host ports cleared, no clash
- Terminal
- Resize, shell detection, no respawn
Every change on the record
Every mutating action, every terminal session and every refused request is in the audit log with who did it. Roles are per environment, and a role can be limited to some namespaces on Kubernetes or some stacks on a swarm. All of it is free on the Community plan.
- Viewer
- Reads lists, logs, stats, YAML
- Operator
- Also changes things, opens terminals
- Admin
- Users, environments, access
One UI past one host.
Add remote Docker hosts, Docker Swarms and Kubernetes clusters next to the local socket. Each gets its own colour and tag, so you always know where you are before you press restart.
prod-euDockerLocal socket12 runningedge-02DockerAgent, dialled out4 runningstaging-k8sKubernetesIn-cluster agent31 podsswarm-prodSwarmAgent per node5 nodeslabDockerPodman socket3 runningRemote hosts run fleetops-agent, which dials out to the server. The host opens no port and never exposes its Docker socket. Logs, terminals, stats and image builds work there exactly as they do locally.
Kubernetes clusters connect by kubeconfig or an in-cluster agent. Lists follow watch streams; every YAML edit is a dry run shown as a diff before anything is written. Helm releases install, upgrade, roll back and uninstall with the same review step.
Docker Swarms run the agent on every node, each proving its node with the node's Swarm certificate. Services, tasks and nodes are live, and stacks deploy as Swarm services.
Made for the teams that run the hosts.
Five situations FleetOps was designed around.
- Shared Docker hostsA login and a role for everyone, instead of SSH and the socket.
- Platform teamsDocker hosts, swarms and Kubernetes clusters under one set of roles.
- On-call and incidentsThe failing container rises to the top. Logs are one key away.
- Edge and remote sitesHosts behind NAT join by dialling out. Nothing is exposed.
- Air-gapped and regulatedNo call home, offline licenses, an audit log you can export.
The same afternoon, from the command line.
FleetOps does not replace the CLI. It saves you from typing it under pressure.
| Step | With the docker CLI | With FleetOps |
|---|---|---|
| Find what broke | docker ps -a | grep -v Up | It is already at the top, under Needs attention. |
| Read why | docker logs --tail 500 -f orders-api | Enter. 200,000 lines, regex search, nothing repeated on reconnect. |
| Fix it safely | docker inspect, docker rm, docker run … by hand | Recreate with the change. A failed start rolls back to the old one. |
| Say who did it | Scroll the incident channel | In the audit log, with the user, the target and the outcome. |
A socket is root. FleetOps treats it that way.
Every touch of a Docker host is signed in, allowed by a role and written down.
- Signed in, every route
- Sessions and CSRF on every Docker call; the socket is never anonymous.
- Agents dial out
- Remote hosts open no port and never expose their socket.
- Written down
- Every change, terminal session and refused request, with who did it.
- No call home
- Offline licenses, no telemetry, a distroless non-root image.
Weight is a feature.
A Rust binary with the web UI and a SQLite database inside. Nothing else to run, nothing to upgrade in lockstep.
| Server | One static binary, about 45 MB | linux/amd64, linux/arm64 |
|---|---|---|
| Image | About 20 MB compressed per platform | distroless, runs as non-root |
| Storage | One SQLite file in WAL mode | no external database |
| Memory target | Under 50 MB idle with about 50 containers | a design target, not a benchmark |
| Heavy parts | Terminal and charts load only when opened | per tab, on demand |
Free to run. Priced by the node when you grow.
One image. Community has no node or user limit; a license key turns on image scanning, CVE alerts, SSO and the other Enterprise parts.
Community
Free
The whole product, for any team, forever.
Unlimited nodes and users
Starter
$9/ month for 10 nodes
Small teams that want the Enterprise parts.
10, 20 or 30 nodes
Scale
Recommended$19/ month for 10 nodes
Teams running FleetOps in production.
10 to 50 nodes
Enterprise
Custom
Larger fleets and procurement needs.
More than 50 nodes
Yearly billing: Starter $99, Scale $199 per 10 nodes. Full pricing
From the blog.
Release notes, and how the parts of FleetOps work inside.
- FleetOps 1.3: run the server in Kubernetes with HelmA Helm chart and a plain manifest for running FleetOps in your own cluster, with its database on a kept volume and HTTPS for agents.
- FleetOps 1.4: manage the cluster FleetOps runs inInstalled in Kubernetes, FleetOps registers its own cluster as the local environment through its service account, and a new start page leads straight to adding an environment.
- FleetOps 1.1: CVE notificationsEnterprise rules send new vulnerabilities in the images you run to Slack, email, Microsoft Teams, Discord, PagerDuty or a webhook: one message per CVE, batched, with no flood when you start.
See your own containers in a minute.
docker run -d --name fleetops -p 9000:9000 \
-v fleetops-data:/data \
-v /var/run/docker.sock:/var/run/docker.sock \
--group-add "$(docker run --rm -v /var/run/docker.sock:/s busybox:1.37 stat -c %g /s)" \
fleets/fleetops