Version 0.11.0-rc

​

What’s new in v0.11.0-rc?

​

✨ New features & improvements

• Kubernetes support: Collabase now ships an official Helm chart. If your team runs a Kubernetes cluster, you can install, upgrade, and back up Collabase with standard Helm commands — see the new Kubernetes installation guide.
• Docker Swarm support: A ready-made Swarm stack file with managed secrets, rolling updates, and automatic rollback — see the new Swarm installation guide.
• Installer as a management tool: install.sh now understands commands — status, logs, stop, start, restart, restore, support-bundle, and uninstall — plus --help on everything. The updater gained --dry-run (preview an update without changing anything), --yes for unattended updates, version pinning (--version v0.12.0), and a rollback command that returns to the previously running version.
• S3 object storage (optional): Uploads and attachments can now live in any S3-compatible bucket — AWS S3, MinIO, Cloudflare R2, or Infomaniak Object Storage — instead of the local disk. Local disk remains the default and needs no changes. A migration command copies existing files into the bucket. See the new File storage guide.
• Horizontal scaling on Kubernetes: With S3 storage and the new dedicated worker option, the Helm chart now supports running multiple application instances. Single instance remains the default.
• Simpler AI setup: AI features (Brain, chat, AI automation steps) now run inside the application and use the provider you pick under Settings → AI. There is no separate AI service anymore, no extra RAM tier, and no auto-detection — every server gets the same installation.

​

🔄 Changes

• One installation profile: The previous “with AI” / “without AI” split based on server memory is gone. 2 GB RAM is the minimum, 4 GB is recommended, and all features are available on every installation.
• AI provider is no longer preselected: New installations start with no AI provider chosen. Pick one under Settings → AI — the previous default pointed at a local Ollama server that most installations did not have.
• Removed: external MCP tool connections. The option to call tools on external MCP servers from automation workflows has been removed, along with its settings. Automation AI steps (prompt, chain, agent) continue to work and now use your configured AI provider. If your workflows used MCP tool steps, those steps will show as failed after the update — remove or replace them.

​

🛠 Stability & Maintenance

• Cleaner updates: The next update automatically removes the no-longer-needed AI engine service and tidies obsolete configuration entries. No manual steps required.
• Faster, more reliable startup: The application no longer downloads tooling from the internet while starting — everything it needs ships inside the image. Installations without internet access at runtime are now supported.