What's new

• Anyone can sign up, accepting the terms, acceptable use policy and privacy policy, then confirm their email address.
• Each new workspace waits for a hand review during the beta. Suspended and pending workspaces can't act.
• Every workspace brings its own servers, Coolify, GitHub, DNS, registrars, SMTP and AI keys.
• Accounts can be exported and deleted.
• The public site, with docs, the services catalogue and draft legal pages.

• Launch your stack buys a server in your own Hetzner account (DigitalOcean in testing), after your approval.
• It installs Coolify, finishes setup over SSH so the API token never crosses the internet in clear, and points coolify. on your domain at it.
• The connection is added only once that HTTPS address answers.
• Failed steps can be retried on their own; nothing is bought twice.
• An owner can reveal Coolify's admin password once.

See Launch your stack.

• All 352 one-click services from Coolify's template feed, kept as a snapshot and refreshed daily in live mode.
• Logos are copied and cleaned, and served by airepeat.
• Services that mount the Docker socket, run privileged, use host networking, publish host ports or open an unauthenticated admin page need the owner's acknowledgement.
• Services can be started, stopped, restarted and given variables.

Browse the catalogue.

• Keys typed into the dashboard are stored with AES-256-GCM under a rotatable master key, never shown again and audited by name only.
• AI keys, alert SMTP, GA4, GitHub webhook secrets and preview passwords are per workspace.
• Requests to addresses you choose resolve once, refuse private, loopback, link-local and metadata addresses, and connect to exactly the address checked.
• Rate limits and local content checks on sign-up, AI work and other actions, and a public abuse report form.

Read How airepeat keeps your keys safe.

• Claude is used on real sites today. The other providers are built in and tested, but not yet run against real accounts.
• Built in: Claude, OpenAI's GPT-6 models and Gemini 3.x.
• Added by name per workspace: DeepSeek, Kimi, OpenRouter and Ollama models.
• Every model picker shows only the models whose provider is set up, with effort where the model takes it.
• Usage is measured the same way for every provider, so cost limits and the monthly AI budget apply to all of them.

• Stopping or rebooting a server needs an owner's approval, because every site on it goes down.
• Starting a server, and starting, stopping or restarting apps and databases, run straight away for people with the right role, and are audited.
• Each server, app and database has its own page with details, usage, related sites and recent operations.
• Provider data refreshes automatically every 15 minutes by default, for every connected provider.

• WordPress: a WordPress site is Coolify's one-click WordPress with MariaDB. Coolify generates its passwords; airepeat never reads them.
• GitHub import: pick a repository and airepeat reads its branches and files, without running them, to suggest the build (Dockerfile, Compose or Railpack), port, health path and memory.
• The chosen branch now reaches Coolify.
• Automation templates: nightly backups, weekly retention, alert checks and more, each listing exactly what it creates.

• Create with AI lists your past conversations to resume, view or archive.
• Voice input uses the browser's own speech recognition. airepeat sends and stores no audio, and hides the button where the browser doesn't support it.
• Long chats scroll in place and follow new text unless you've scrolled up.
• Brand icons for providers, a page loading bar, and a sidebar you can resize.

• Files: browse a site's code and edit it on your own draft branch, then send it to review.
• Live logs: follow an app's output as it happens, with credentials and email addresses masked.
• SEO: settings live in seo.json in the site's repository, with a check that crawls up to 50 pages.
• Variables: change an app's variables through an approval; secret values are never stored.
• Monitoring: real sites show only real readings, plus TLS certificate expiry. Alerts can go to Slack and signed webhooks.

• New templates: Astro, Nuxt, SvelteKit, Express, Laravel, Symfony, Django, FastAPI and Go.
• The Next.js templates move to Next.js 16, and new sites keep only the template parts their plan needs.
• A site's database can be PostgreSQL, MySQL or MariaDB, and a site can add Redis.
• Every app gets its template's memory limit, so one runaway site can't take the server down.

• A full-stack template with a database, an admin area, stored forms and a simple shop.
• Each full-stack site gets its own private Postgres database, after approval.
• Payments (Stripe, test mode only), email over any SMTP account, and file storage on your own Cloudflare R2 bucket.
• Secret keys are typed in once and handed to Coolify from memory, never kept.
• Previews get their own empty database, so they never touch production data.

• Daily database backups at 03:00 UTC, kept by Coolify for 7, 14 or 30 days, with optional copies off the server.
• Health checks every 5 minutes after launch. Three failures in a row mark a site degraded.
• A failed deploy offers Ask AI to fix, with the end of its build log. The fix goes through review like any change.
• Alert emails through an SMTP account you connect.

See Backups and Monitoring and alerts.

• Set up preview gives a site a second app at a hidden address, protected by a password.
• AI changes are built and deployed to the preview before review, so you can check them first.
• Previews never share production's secrets.

• A request still waiting for approval now counts in every period until it settles, not just the month it was made in.
• Before, around the turn of a month, up to about twice a limit could be committed.
• This covers the monthly spending limit and the daily limits on new servers and domain registrations.

• AI tasks work on their own copy of the site's repository, on their own branch.
• The platform, not the model, enforces the tool policy: nothing outside the repository and no symbolic links.
• You can steer a task between steps.
• AI work reaches production only through the release review.

• Five roles: owner, administrator, developer, operator and viewer, checked on the server for every page and action.
• One engine for every provider change, with approvals bound to the exact resource and parameters, and idempotent retries.
• Simulation by default: nothing reaches a provider unless live mode is switched on.
• Releases bind a review to the exact code and configuration; a launch gate guards production.
• Domain purchases approve an exact price, and DNS plans never overwrite records silently.
• Every backup airepeat takes is restored before it counts, and sites are archived before they can be purged.