Self-hosted monitoring for homelabs: host and container metrics and uptime checks from all your machines in one dashboard, with an optional AI agent that diagnoses incidents and suggests fixes.
Python ai-agentsaiopsdashboarddockerfastapihomelabincident-managementllmmonitoringopenaireactroot-cause-analysisself-hosteduptime-monitoring
Changelog
All notable changes to Infradeck are listed here. The format follows Keep a Changelog, and versions follow Semantic Versioning.
1.0.0 - 2026-11-12
The first public release. Earlier 0.x versions were private pre-releases.
What's in it
- Hosts and containers. CPU, memory, load, swap, network and uptime for every server, disk and inode usage for every mounted filesystem, and per-container CPU, memory, I/O, ports, health, exit codes and history.
- System events. Reboots, OOM kills, kernel faults, clock sync, pending reboots and package updates. With the optional system helper, also clean or unclean shutdowns, which process the OOM killer stopped and who listens on each port.
- Network. Listening ports (and which ones other machines can reach), firewall changes and an optional internet check.
- Statistics for any range, from the last hour to all time, per server, per container and for the whole fleet, kept fast by hourly, daily, weekly and monthly rollups. Includes the busiest container over time, CPU spikes and a "disk full in about N days" estimate.
- Availability checks run from the servers you pick: HTTP(S), a keyword, a value in a JSON reply, a TCP port, TLS certificate expiry, a DNS record and ping. They have retries, an "all fail" or "any fails" rule, upside-down mode, a Test button, and uptime history for 24 hours, 30 days and a year.
- Incidents. One incident per problem, opened after a sustained breach and closed when it clears. Admins can edit thresholds and switch any rule off; incidents can be acknowledged, muted and resolved.
- Notifications to ntfy, Discord, any webhook or email, with per-severity routing and quiet hours.
- Incident agent (optional). With an Anthropic, OpenAI or Gemini API key, it diagnoses warning and critical incidents from masked evidence, looks things up with read-only tools within a budget, cites the data behind each claim, proposes a fix and watches whether it holds. You can also chat about any incident. It never runs commands. Monthly budget, per-model prices and who may chat are set in Admin.
- Accounts. An admin and read-only users, invitations, two-factor sign-in (required for admins), sessions you can review and end, and read-only or write API tokens.
- Adding servers. One command and one click: a join token, then approval on the dashboard. Collector images are signed with cosign and pinned by digest.
- Published images. Every Infradeck image (backend, frontend, collector, Docker proxy, system helper) is
built by the release workflow for amd64 and arm64, signed with cosign and published on GHCR. A server install
is a checkout of the release tag plus
docker compose pull && docker compose up -d; nothing is built on the server unless you adddeploy/build.yml.
Security
- Collectors never get the Docker socket. A small proxy allows only a few read requests and strips what the collector doesn't need.
- With
docker-compose.server.yml, collector endpoints exist only on a separate gate on your VPN address, and the backend refuses them anywhere else (COLLECTOR_ACCESS=strict, now also the backend's default). - The login cookie is
Secureby default (SESSION_COOKIE_SECURE=true). Sessions are stored hashed, and secrets in the database (two-factor keys, the agent's API key, notification targets) are encrypted. - Turning the agent on, choosing its provider and key, and letting a user chat with it need the password and a two-factor code.
- Everything sent to the agent's model provider is masked first, including chat messages.
See SECURITY.md for the full picture.
Upgrading from a 0.x pre-release
Back up the database first, then follow Updating. Changes since 0.10.0 that may need your attention:
- License. Infradeck is now licensed under the AGPL-3.0 (it was MIT).
- Traefik labels moved to
deploy/traefik.yml. SetCOMPOSE_FILE=docker-compose.server.yml:deploy/traefik.ymlin.envbefore updating, or Traefik stops serving the dashboard. - The frontend listens on port 8080 inside its container and runs without root. A reverse proxy you set up
yourself to reach
frontend:80must usefrontend:8080. TRUSTED_PROXIES.X-Forwarded-Foris trusted only from the addresses you list there. Set it to your reverse proxy, or every client shares the proxy's address for rate limits.- The incident agent's provider, model and API key are chosen in Admin → Incident agent.
OPENAI_API_KEYandAGENT_MODELin.envare no longer read: enter the key there again, then remove them from.env. - Backend defaults.
SESSION_COOKIE_SECUREnow defaults totrueandCOLLECTOR_ACCESStostrict. The compose files set what they need; only a backend run some other way over plain HTTP must set them itself. - Metrics ingest accepts only a server's collector token, not a signed-in session.
GET /dashboard/summarywas removed.- Collectors. Update them with a freshly generated command to get hard time limits on checks and the refusal to follow a redirect from the dashboard's address.
- Images are pulled, not built. Check out the tag (
git fetch --tags && git checkout v1.0.0) and rundocker compose pull && docker compose up -dinstead ofgit pulland--build. Your volumes stay as they are. To keep building on the server, adddeploy/build.ymltoCOMPOSE_FILE. Afterwardsdocker image pruneremoves the old localinfradeck-*images.