Huginn is yours: the config file, the database, the running processes. Your vibe-code agent can edit all of it from the chat and restart the stack when a change needs one. These ten customizations cover setup, tidy data, and hardening, and most take minutes.
1. Make “7am” Mean 7am
Huginn schedules run in the server’s time zone, and containers default to UTC. That means your “7am digest” quietly fires at odd hours until the zone is set. Have the agent set the time zone in the .env file (or config/application.rb if your build ignores it), restart Huginn, and then watch one scheduled agent fire when you actually expect it. This is the fix that makes every other schedule in this list trustworthy. Say: “Set Huginn’s time zone to Europe/Berlin and make sure my 7am digest fires at 7am local.”
2. Real Email Delivery
The EmailAgent silently does nothing until SMTP is configured: server, port, username, password, authentication mode, and a From address. Give the agent your mail server’s details and ask for a test message from a throwaway EmailAgent before you wire real alerts to it. Most mail servers work on port 587 with STARTTLS or 465 with SSL, and the agent can try both. Say: “Configure Huginn to send email through my SMTP server and prove it with a test message.”
3. Fix the Domain
Links inside alert emails and digest items are built from the DOMAIN variable in .env. Set DOMAIN to your real address so digests link back correctly when you read them from a laptop outside the server. Without it, links point at an internal address that resolves to nothing from anywhere else. It is a one-line change and a restart. Say: “Set DOMAIN so the links in my alert emails point to https://huginn.mydomain.com.”
4. Standard Event Shapes
Every agent emits JSON events, and without rules the keys drift: one source says url, another says link, a third nests everything under data. Have the agent place an EventFormattingAgent or FormatterAgent after each source so every event looks like {source, title, url, occurred_at}. The FormatterAgent can also rename keys, flatten nested JSON, and drop fields you never use, which keeps your event list readable. Downstream agents stop needing per-source hacks, and your digests format cleanly. Say: “Normalize every incoming event to source, title, url, and occurred_at, whatever the source was.”
5. Dedupe Keys That Actually Work
The DeDuplicationAgent needs one stable property per source to compare events on: the posting URL for jobs, an id field for API items. Pick the wrong key and you get repeats; pick the right one and duplicates vanish silently. Ask the agent to inspect a few real events from each source and choose the key with evidence, showing you two events side by side so you can see it working. Say: “Dedupe my job alerts on the posting’s URL so I never see the same role twice.”
6. Credentials Out of Agent Options
API keys and webhook URLs pasted into agent options end up in exported scenario JSON, which makes sharing or backing up scenarios a leak. Move them into Huginn’s Credentials store and reference them in options as {% credential slack_webhook %}. Do this before you export or share any scenario, because it is much harder to scrub a key after it has been copied. Rotating a key becomes editing one row instead of hunting through every agent. Say: “Move my Slack webhook URL and API keys into credentials, and rewrite the agents to use them.”
7. Scenario Hygiene With Version Control
Scenarios group related agents, and they export to portable JSON. Have the agent impose a naming convention ([watch] competitor, [alert] uptime), keep one scenario per purpose, and dump every scenario to JSON into a git repo on the server each night. You can diff changes, roll back a bad edit, and copy workflows to a second server as a paste job instead of a rebuild. Say: “Export all my scenarios to JSON in a git repo every night, and rename everything with a [watch]/[alert] convention.”
8. Keep the Database Small
Huginn keeps every event, and the events table is what grows without limit. Set EVENTS_EXPIRE_AFTER in .env to a few days: your digests have already rolled up what matters, and long history belongs in the digests, not the database. The default retention is already a week, so shorten it only if you poll heavy sources like search results. A small database means a fast UI and fast backups. Say: “Expire events after 7 days and prune old agent logs so the database stops growing.”
9. Backups That Restore
Everything Huginn knows lives in one MySQL database: agents, credentials, scenarios, and recent events. A nightly mysqldump cron with fourteen days of rolling files is enough for most servers. Treat the dumps like the database itself, with the same permissions and backup rules, because they contain your credentials. Then do the part everyone skips: once, restore the dump into a scratch database and confirm the agents look right, so you know the backup is good before you need it. Say: “Set up a nightly database dump with 14-day rotation, and restore it into a scratch database once to prove it works.”
10. Tighten the Surface
Huginn does not need a public port. Read alerts by email or RSS, and reach the web UI through an SSH tunnel or a reverse proxy that adds its own auth. Rotate APP_SECRET_TOKEN, put long random paths on every WebhookAgent and DataOutputAgent endpoint, and keep the admin password unique to this server. If you do put it behind a proxy, the agent can add the forwarding headers Huginn expects so redirects keep working. Ask the agent to check the firewall rules while it is in there. Say: “Check that port 3000 is not exposed publicly, rotate APP_SECRET_TOKEN, and give my webhook endpoints long random paths.”
Most of these are one-time asks with a permanent payoff. The honest ordering: email and time zone first, credentials and dedupe once you have real agents running, backups before the server matters to you. If a task above touches the database, ask the agent to take a dump first, that habit turns every risky change into a reversible one.
Huginn ships preinstalled on OpenSysLab servers, agent included. Huginn on OpenSysLab
