Setup work and ops hardening are the jobs nobody wants to do by hand — which is why half of self-hosted Plane installs run on defaults. Your vibe-code agent can edit the environment files, restart the Docker services, run Django shell commands inside the api container, and query the database directly. Here are ten customizations to ask for, ordered from “ten minutes” to “an afternoon”. Each includes the exact sentence to type.
1. Design your states pipeline
Plane lets every project define its own states, grouped into backlog, unstarted, started, completed, and cancelled. The agent can create the states your process actually uses — In Review, QA, Ready for Deploy — set the default state new issues land in, order them on the board, and pick which ones count as closed for reporting. Getting the groups right matters more than the names, because analytics aggregate by group, not by state. Say: “For project WEB set up states: Backlog, Ready, In Progress, In Review, QA, and Done — Done is the only completed state, and new issues start in Backlog.”
2. Set the estimate scale
Plane supports both point estimates and time estimates, per project, and the scale should match how your team thinks. The agent switches the projects that want points to a scale like 1, 2, 3, 5, 8, 13, keeps time estimates where they fit better, and verifies existing estimates survive the change untouched. Changing the scale mid-project skews velocity history, so ask it to note the switch date in a page for future reference. Say: “Switch WEB to point estimates with the scale 1,2,3,5,8,13 and make sure existing estimates stay intact.”
3. Define a label taxonomy with colors
Labels only work if they mean the same thing to everyone. Have the agent design a convention — green area labels (area-billing, area-auth), red type labels (type-bug, type-feature), and one release label per version — then rename and recolor what exists and fill the gaps. Consistent colors make boards scannable at a glance, and the naming convention keeps new labels from sprouting randomly later. Say: “Give me a label scheme — area labels in green, type labels in red, and one release label per version — then rename what we have to match and fill the gaps.”
4. Add Intake routing rules
Your Intake queue is more useful when issues arrive pre-sorted. The agent writes a small scheduled job that scans pending intake issues and applies keyword rules: anything mentioning “payment” goes to Priya with the billing area label, security keywords set urgent priority, performance keywords get tagged for profiling later. You tune the rules in plain language afterward, and the agent updates the script. Say: “Set up intake routing: anything mentioning payment goes to Priya with the billing area label, and anything with security words gets urgent priority.”
5. Set instance name and lock down signups
Out of the box your instance carries a default name and open registration, which is not what you want on a private server. Through the instance admin app (god-mode), the agent sets a proper instance name, turns off open signups, and restricts registration to your email domain so only your team gets in. It then restarts and verifies the settings took effect, so you are not trusting a config file nobody checked. Say: “Change the instance name to Acme Projects, turn off open signups, and allow only @acme.com addresses.”
6. Wire up Google or GitHub sign-in
OAuth login means one fewer password for the team and faster offboarding when someone leaves. The agent walks you through creating the client ID and secret in the provider’s console — the only manual step, about five minutes — then adds the variables to the environment file, restarts the web and api services, and tests a login end to end. Be honest with yourself: skip the console step and nothing will work, and the agent will say so. Say: “Hook up GitHub sign-in: tell me exactly which client ID and secret to create, add them to the env file, restart the services, and verify login works.”
7. Make email actually work
Notifications, mentions, and invites all depend on SMTP, and on most self-hosted installs this is the step that was never finished. The agent sets the mail host, port, credentials, and sender address in the environment, restarts the worker, and proves it by sending a real test mention you receive in your inbox. If it fails, it reads the container logs and tells you whether the problem is auth, TLS, or DNS. Say: “Configure SMTP with our mailbox settings and send a test notification so I can see mention and invite emails actually arrive.”
8. Nightly backups you have watched restore
A backup you have never restored is a hypothesis, not a backup. The agent schedules a nightly pg_dump of the Plane database plus a copy of uploaded attachments, keeps fourteen days, and — the important part — performs a restore into a scratch database while you watch. Expect it to set up local copies first; pushing them off-server via rsync is the follow-up worth asking for the same day. Say: “Back up the database and uploaded files nightly at 2am, keep two weeks of copies, and this week show me a restore into a scratch database to prove it works.”
9. Housekeeping for containers and database
Small self-hosted stacks die quietly from logs that fill the disk and containers with no memory limits. The agent adds memory limits to the api and worker containers, sets log rotation, schedules a weekly Postgres vacuum and analyze, and prunes old Docker images. Nothing here is dramatic; it is the difference between an install that runs for years and one that falls over in month two. Say: “Put memory limits on the api and worker containers, rotate the logs so they can’t fill the disk, and schedule a weekly database vacuum.”
10. Health checks with real alerts
You should find out Plane is down before your team does. The agent adds a cron job that every five minutes confirms the web, space, api, and worker containers are running and healthy, checks disk usage, and emails you the moment anything is off. Ask it to append each check to a log file, so you get uptime history instead of surprises, and to include TLS certificate expiry if your domain is public. Say: “Every 5 minutes, check that web, api, and worker are healthy and disk is under 80%, and email me the minute anything is off.”
Work through these in order and your Plane install ends up shaped to your team instead of shipped defaults. The agent does the typing; you make the calls. To get a private server with Plane and the agent already installed, see Plane on OpenSysLab.
