See the shape of your latency. Not just its average.

Driftband is a self-hosted network latency monitor. It pings your hosts on a schedule and draws smoke graphs: the full distribution of round-trip times as shaded percentile bands, with a median line coloured by packet loss. It runs as one Docker container next to PostgreSQL.

Driftband is inspired by SmokePing, created by Tobias Oetiker in 2001, and keeps its way of showing latency. It is a new implementation written in Go and JavaScript and contains no SmokePing code. It adds what SmokePing never had built in: a web interface for configuration, organizations and roles, two-factor sign-in and an audit log.

Yes. Paste your SmokePing config or upload the file. Driftband reads the *** Targets *** section, recreates the + / ++ / +++ folder and host hierarchy with host, menu and title, and attaches your default probe profile to every host. Probes and alerts are not imported, branches deeper than three levels are flattened, and the import only adds to your tree. Only plain-text configuration files are accepted (up to 2 MB and 1000 hosts per import); binary files, scripts and invalid host addresses are refused before anything is created.

ICMP echo round-trip time and packet loss. Each probe cycle runs fping in count mode and sends several pings to the host; Driftband stores every individual round-trip time, so it can show percentiles instead of a single average.

Yes, with probe profiles: the interval between cycles (5 seconds or more), the number of pings per cycle (1 to 100, and never more than the interval in seconds) and the ICMP packet size (12 to 1472 bytes, which is a 1500-byte Ethernet frame minus the IPv4 and ICMP headers). Every organization starts with Standard (20 pings every 60 s), Fast (every 30 s), Coarse (every 300 s) and Large packet (15 pings of 1400 bytes). You can create your own, mark one as the default and choose a profile per host.

The darkest band holds the middle 50% of pings, the lighter bands the middle 80% and 95%, and the faintest band the full range from fastest to slowest ping. The wider the smoke, the more jitter. The median line is green when nothing was lost and turns yellow, orange and red as loss grows; a strip along the bottom marks every cycle with loss.

Yes. Each organization has its own hosts, probe profiles, retention setting and audit log, fully isolated from other organizations. Owners and admins invite people with single-use links that expire after 48 hours, and can switch anyone except the owner between the admin and member roles at any time.

Passwords are hashed with Argon2id. Users can enable TOTP two-factor authentication with any authenticator app or hardware TOTP key, with replay protection. Sign-in and password confirmations are rate limited, sessions use HttpOnly cookies, and every action, including failed sign-ins and refused permission checks, is written to the audit log. An automated security test suite covers every API endpoint.

Each probe cycle is stored with all of its round-trip times, plus 5-minute and 1-hour rollups for fast long-range charts. On a production instance this costs about 370 bytes per cycle, so one host with one cycle per minute needs roughly 190 MB per year. Each organization keeps 1, 2 or 3 years of history; older data is removed automatically.

Telegram (a bot posting to a chat, group or channel), Slack (an incoming webhook), Microsoft Teams (a Workflows webhook that receives an Adaptive Card) and email over any SMTP server with STARTTLS or TLS, signed in with a password or with OAuth 2.0 client credentials, including a ready preset for Microsoft 365. One rule can notify several channels, and every channel has a test button.

A rule measures packet loss or median latency over the last few probe cycles, so a single bad cycle never counts. The condition must then hold for the number of minutes you set before anything is sent. Recovery uses a lower threshold with its own duration, so a link hovering around the limit does not alternate between alert and recovery. If the state still changes too often within a window, Driftband sends one flapping notice and stays quiet until the link has settled. Gaps without measurements never complete a condition.

Owners and admins of an organization create and edit channels and alert rules; members can see them, the active alerts and the delivery history. Bot tokens, webhook URLs, SMTP passwords and OAuth client secrets are stored encrypted and are never shown again after saving. Every change is recorded in the audit log.

Not yet. Driftband measures ICMP round-trip time and packet loss with fping. If you depend on DNS, HTTP or TCP probes, SmokePing still covers those cases.

Yes. The public test instance at https://test.driftband.app has every feature. Each organization there can have up to 10 hosts, probe intervals of at least 60 seconds and ICMP packets of at most 56 bytes. A self-hosted Driftband has none of these limits; its administrator can set a host limit per organization if needed.

With Docker Compose: run the lukmorawski/driftband image, pinned to a version tag such as 0.1.0, next to PostgreSQL 16, open the web interface and create your organization. The image runs as a non-root user and needs no extra Linux capabilities: fping uses unprivileged ICMP sockets. Put it behind a reverse proxy with HTTPS for production. The compose file above is all you need.

Driftband is free, self-hosted software licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). There is no usage limit and no telemetry. If you run a modified version as a network service, the AGPL asks you to offer its source code to your users.