# docker-compose.yml
services:
maintenant:
image: ghcr.io/kolapsis/maintenant:latest
ports:
# ⚠️ SECURITY: publishes the UI/API (no authentication of their own) on
# every interface; put an auth reverse proxy in front, or bind "127.0.0.1:8080:8080". See https://docs.maintenant.dev/security/#reverse-proxy-setup.
- "8080:8080"
read_only: true
security_opt:
- no-new-privileges:true
tmpfs:
- /tmp:noexec,nosuid,size=64m
volumes:
- /var/run/docker.sock:/var/run/docker.sock:ro
- /proc:/host/proc:ro
- /etc/os-release:/host/etc/os-release:ro
- maintenant-data:/data
environment:
MAINTENANT_ADDR: "0.0.0.0:8080"
MAINTENANT_DB: "/data/maintenant.db"
restart: unless-stopped
volumes:
maintenant-data:
Docker monitoring for AI agents: logs, history, alerts. Can ack alerts; never touches containers.
2026-11-09
ดาว
—
เครื่องมือ
—
อัปไทม์
—
docker compose up -d
Open http://localhost:8080. Your containers are already there, with their health, restart loops, resources and logs. Nothing to configure.
Docker socket access is automatic: the entrypoint reads the mounted socket's group and grants it to the unprivileged user, on Compose and on Swarm (where docker stack deploy silently ignores group_add). A socket owned by group root needs DOCKER_GID=0. If containers do not show up, see Troubleshooting.
Endpoints, certificates and heartbeats work without any container runtime. Container monitoring switches on by itself the moment a runtime shows up. Install documentation for pinned versions, air-gapped installs and supply-chain verification.
Monitoring your own infrastructure with the standard stack means running Prometheus, Grafana, Alertmanager, node-exporter, cAdvisor, blackbox-exporter, a certificate exporter, Loki, Promtail, Trivy and something for image updates. Ten-odd components, each with its own config, upgrades and dashboards, to answer one question: is my stack up, and what is burning?
maintenant answers that question with one container.
Built into maintenant
What you would assemble instead
Container state, health checks, restart loops
cAdvisor + node-exporter + alert rules you write
CPU, memory, network and disk, per container and per host
cAdvisor + node-exporter + Grafana dashboards you build
HTTP / TCP endpoint checks, declared as Docker labels
blackbox-exporter + a config file per target
TLS certificate expiry and chain validation
ssl_exporter
Cron and heartbeat deadlines
Pushgateway + alert rules you write
Live container logs, stdout/stderr demuxed
Dozzle, or Loki + Promtail
Image update detection, with compose-aware update and rollback commands
CVE enrichment and per-container risk score (Personal)
Trivy + its exporter
Alerts routed to Discord, webhooks, email, Telegram, Slack and Teams, with escalation (Pro)
Alertmanager
Public status page with incidents and subscribers
Cachet, Uptime Kuma, or a SaaS
One real-time dashboard for all of the above
Grafana + dashboards you build and maintain
Do I still need Prometheus?
maintenant monitors your infrastructure. Prometheus monitors your application. There is no PromQL here, no custom exporters, no panels to design: maintenant already knows what a container, a certificate, an endpoint, a cron job and a CVE are, and starts watching them the moment they appear. If you ship business metrics and write your own queries, keep Prometheus for that. The two answer different questions, and plenty of people run both.
Against the tools usually stacked up next to it:
maintenant
Uptime Kuma
Portainer
Dozzle
Container auto-discovery
Yes
No
Yes
Yes
MCP server, agent-ready
Built in
Third-party
Separate
No
Live container logs
Yes
No
Yes
Yes
HTTP/TCP endpoint checks
Yes
Yes
No
No
Cron/heartbeat monitoring
Yes
Yes
No
No
SSL certificate tracking
Yes
Yes
No
No
CPU/memory/network metrics
Yes
No
Limited
No
Image update detection
Yes
No
Yes
No
Network security insights
Yes
No
No
No
CVE enrichment, risk scoring
Personal
No
No
No
Public status page
Yes
Yes
No
No
Alerting with routing
Yes
Yes
Limited
No
Kubernetes native
Yes
No
Yes
No
Single binary, zero deps
Yes
Node.js
Docker API
Docker API
Runs without a runtime
Yes
No
No
No
One container. One dashboard. Everything monitored.
Screenshots
Agents
maintenant watches, your AI agent investigates. The loop has three steps:
Detect. Fixed rules decide that something is down: consecutive failures, thresholds, missed deadlines, restart loops. No model is involved, so an alert is never invented and nothing is spent while everything is fine.
Push. The alert leaves as a webhook: alert.fired when it starts, alert.resolved when it recovers, with the host it belongs to (agent_id) and a link back to it.
Investigate. Your agent receives the alert and queries maintenant's built-in MCP server: the container's logs, the endpoint's check history, CPU and memory, the other active alerts on the same host. You get a diagnosis, not just a red light.
flowchart LR
S[containers, endpoints,<br>certificates, cron jobs, hosts] --> M[maintenant<br>rules fire an alert]
M -- webhook --> A[your agent<br>Claude Code, OpenCode, n8n]
A -- MCP: logs, history,<br>active alerts --> M
With Claude Code. A small receiver (Python, standard library only) runs on the host next to maintenant. Each fired alert starts a headless Claude Code session that can only call maintenant's read-only MCP tools, and the diagnosis lands in reports/<alert-id>.md:
# what the receiver runs for each alert
claude -p "maintenant just fired this alert: {...} Investigate it read-only." \
--setting-sources project --mcp-config mcp.json --strict-mcp-config --tools "" \
--allowedTools mcp__maintenant__list_alerts mcp__maintenant__get_container_logs \
mcp__maintenant__get_endpoint_history mcp__maintenant__get_resources
mcp.json reaches maintenant over stdio with docker exec -i maintenant /app/maintenant --mcp-stdio, so no MCP port is opened: start maintenant with MAINTENANT_MCP_SOCKET=true to serve the local socket it relays to. Point a webhook channel at the receiver with an Authorization: Bearer <token> header and route alerts to it with a trigger. The AI agents guide walks through the setup.
With OpenCode, the same receiver calls opencode run instead of claude -p, with maintenant declared as a local MCP server in opencode.json. With n8n, a Webhook trigger feeds an AI Agent node whose MCP Client tool points at your instance's /mcp endpoint.
Built-in Model Context Protocol server with 51 tools. Ask your AI assistant what is burning, read a container's logs, check the alert queue, acknowledge an alert, open an incident. stdio and Streamable HTTP transports, OAuth2 with a client id and secret for remote clients (Claude web, mobile and Desktop).
Zero-config auto-discovery for Docker, Docker Swarm and Kubernetes. Every container is tracked the moment it starts: state changes, health checks, restart loops, live log streaming with stdout/stderr demux. Compose projects are grouped automatically. Read-only: maintenant observes, it never touches your containers.
One central server, lightweight read-only agents on your other hosts, a persistent mutually-authenticated gRPC stream between them. No shared database, no message queue, no PKI to run: an agent enrolls with a one-time token and an Ed25519 keypair generated locally, and every stream is challenge-response authenticated. Revoke it from the UI at any time.
# On each remote host, one command, generated for you in the UI
docker run -d --name maintenant-agent --restart unless-stopped \
-v /var/run/docker.sock:/var/run/docker.sock:ro \
-v /proc:/host/proc:ro \
-v /etc/os-release:/host/etc/os-release:ro \
-v maintenant-agent-data:/var/lib/maintenant \
ghcr.io/kolapsis/maintenant:latest \
--mode=agent --server=grpcs://monitoring.example.com \
--enrollment-token=mnt_enr_XXXXXXXXXXXXXXXX --label="prod-worker-01"
Agents detect their local runtime (Docker, Swarm or Kubernetes), stream container state, endpoints, certificates, host CPU/memory/disk, and reconnect on their own, replaying what they saw while disconnected as history. Every entity is attributed to its host, so nothing gets mixed across machines. Personal: up to 20 remote machines. Pro: unlimited.
Scans OCI registries: newer versions for fixed tags, and for floating tags like latest a comparison between the digest your container runs and the one the tag points at now. You know which images have an update before you docker pull blindly. Update and rollback commands for Compose, plain Docker, Swarm and Kubernetes, with the right cd into the Compose project. No Diun, no Watchtower, no extra container: it is part of the monitor.
Every monitored host reports its distribution and version; maintenant checks it against the support cycles of Debian, Ubuntu, RHEL, Rocky, Alma, Alpine and SLES and alerts 30 days before the free security support ends, then again once it has. Dates ship with the binary and refresh daily from endoflife.date when the network allows.
HTTP and TCP checks declared as Docker labels, picked up when the container starts, or added by hand. Endpoints can also be derived from Traefik and Caddy labels (opt-in). Response times, uptime history, 90-day sparklines, failure and recovery thresholds.
Create a monitor, get a URL, add one curl to the job. maintenant tracks start and finish, duration, exit code, and alerts when the deadline is missed. Outbound heartbeats let two maintenant instances watch each other, so a dead monitor does not go unnoticed.
Auto-detected from your HTTPS endpoints, plus standalone monitors for any domain. Full chain validation, alerts at 30, 14, 7, 3 and 1 day before expiry, OCSP stapling checks (Personal).
Real-time CPU, memory, network and disk I/O per container and per host, top-consumers view for instant triage, per-container thresholds with debounce. History: 7 days on Community, 30 on Personal, 90 on Pro.
Flags what should not be there: ports bound to 0.0.0.0, exposed database ports, host-network mode, privileged containers, Kubernetes Services of type NodePort or LoadBalancer, and database ports exposed through them. Personal adds CVE enrichment (each image is mapped to its software ecosystem through OCI manifest inspection), a risk score per container and a unified security posture dashboard.
One alert pipeline for every source: container restart loops, unhealthy checks and stopped containers, endpoint failures, missed heartbeats, expiring or invalid certificates, CPU and memory thresholds, available updates, Swarm and Kubernetes health, agents going offline, hosts whose OS loses support. Channels are silent by default and routed through triggers (severity, source, scope). Alerts can be acknowledged. Silence rules for planned maintenance, three delivery attempts per notification.
Channels: Discord and webhooks (Community), email and Telegram (Personal), Slack and Microsoft Teams (Pro). Pro adds escalation policies of up to five levels that page the on-call, then the backup, then the lead, plus maintenance windows.
Real-time status page with severity aggregation across every monitor, live over SSE. Personal adds incident timelines, Pro adds email subscribers (double opt-in, through your own SMTP server), maintenance windows and branding.
Configuration
Everything is driven by Docker labels and a handful of environment variables. No YAML to maintain.
REST API under /api/v1/, plus an SSE event stream.
services:
maintenant:
image: ghcr.io/kolapsis/maintenant:latest
ports:
# ⚠️ SECURITY: publishes the UI/API (no authentication of their own) on
# every interface; put an auth reverse proxy in front, or bind "127.0.0.1:8080:8080". See https://docs.maintenant.dev/security/#reverse-proxy-setup.
- "8080:8080"
read_only: true
security_opt:
- no-new-privileges:true
tmpfs:
- /tmp:noexec,nosuid,size=64m
volumes:
- /var/run/docker.sock:/var/run/docker.sock:ro
- /proc:/host/proc:ro
- /etc/os-release:/host/etc/os-release:ro
- maintenant-data:/data
environment:
MAINTENANT_ADDR: "0.0.0.0:8080"
MAINTENANT_DB: "/data/maintenant.db"
api:
image: myapp:latest
labels:
maintenant.group: "production"
maintenant.endpoint.http: "http://api:3000/health"
maintenant.endpoint.interval: "15s"
maintenant.alert.severity: "critical"
postgres:
image: postgres:16
labels:
maintenant.endpoint.tcp: "postgres:5432"
maintenant.alert.severity: "critical"
redis:
image: redis:7-alpine
labels:
maintenant.endpoint.tcp: "redis:6379"
volumes:
maintenant-data:
Security model
No built-in authentication, by design. Like Dozzle and Prometheus, maintenant sits behind your reverse proxy and auth middleware (Traefik or Caddy, Authelia or Authentik). /ping/{token} and /status/ are meant to stay public. Reverse proxy setup.
Read-only everywhere. Docker socket mounted :ro, read-only RBAC on Kubernetes, read-only agents. maintenant never starts, stops or modifies a container. A socket proxy is supported if you would rather not mount the socket at all.
Hardened container. Drops to nobody (uid 65534) once it has fixed the ownership of its volume, read_only root filesystem, no-new-privileges.
Anonymous, opt-out telemetry. One snapshot per hour: counts of monitored objects, edition, storage engine, version and runtime figures (OS, architecture, CPU cores, memory). No hostnames, IPs, names, URLs or keys, ever. MAINTENANT_DISABLE_TELEMETRY=1 turns it off with no background goroutine and no outbound packet. Exact payload and details.
Architecture
One binary, three modes. Go backend, Vue 3 frontend embedded via embed.FS, SQLite inside. The same file runs embedded (single host, default), server (central ingestion) and agent (remote host).
Zero dependencies. SQLite is the only datastore. No Redis, no queue, nothing to administer. A fleet operator may back the server on a PostgreSQL they already run; agents always stay on SQLite.
Runtime optional. Endpoints, certificates and heartbeats run without any Docker socket or Kubernetes API. Container monitoring resumes on its own when a runtime becomes reachable.
Real-time. SSE pushes every state change to the browser and to the status page instantly.
Light on memory. A freshly started instance fits in a few tens of MB; plan on about 100 MB after several days on a busy host. Runs on a Raspberry Pi, a €4 VPS or a NAS.
Community is free forever and runs production infrastructure every day: it is the full product on a single host, not a crippled trial. Personal is bought once. Pro is what a team needs.
Personal covers one person on infrastructure they own or run for themselves, freelancers included, and ships with one year of updates (then €59 per extra year; every version released inside a paid year stays licensed for life). Pro adds the right to monitor other people's infrastructure. Volume pricing and custom agreements: [email protected].
Paid editions are the same binary, self-hosted the same way. The key is verified against the license server and the signed answer is cached: the instance falls back to Community only after 60 days without reaching the server, so a few weeks offline change nothing. Your monitoring data never leaves your infrastructure.
MAINTENANT_LICENSE_KEY=your-license-key # Personal or Pro, restart, done
Support the project
maintenant is built by one developer in Bordeaux, France. No VC, no ads, no acquisition exit. 100% of revenue funds full-time development. Ranked by impact:
Buy a licence. Personal if the infrastructure is yours, Pro if you run it for others. Unlocks features and pays for the roadmap. See editions →
Sponsor. Any amount, one-off or monthly, credited below. GitHub Sponsors →
Spread the word. Star the repo, share on HN, Lobsters, Reddit or LinkedIn. Discoverability is oxygen for indie projects.
Tell me how you use it. Two minutes, read by the developer, quoted only with your permission. Give feedback →
Backers
Every Personal owner, Pro subscriber and GitHub sponsor keeps this project independent. Thank you.
Code contributions are welcome. Open an issue first for bigger changes; small fixes, typos and docs, just send the PR.
License
Copyright 2025-2026 Benjamin Touchard / kOlapsis, Bordeaux, France.
The core is licensed under the Apache License 2.0. The code that Personal and Pro keys unlock, under internal/commercial/ and frontend/src/commercial/, is source-available under the Maintenant Commercial Source License: free to read and contribute to, production use requires a matching licence key. See NOTICE and COMMERCIAL-LICENSE.md.