Dependabot Has No Path to Forgejo or Gitea, and Only a Community Patch for GitLab: Why Renovate Is the Only Bot Built for All Three
Dependabot has no official route onto GitLab, and none at all onto Gitea or Forgejo. Renovate runs natively on both, plus Bitbucket and Azure DevOps, covers roughly three times the package ecosystems, and ships a dependency dashboard and automerge Dependabot still doesn't have. What that gap actually costs a self-hosted, git-agnostic PaaS.
Rivet Actors Give an AI Agent State That Survives Between Tool Calls — What a Container-Only PaaS Is Missing
Rivet Actors is an open-source, self-hostable take on Cloudflare Durable Objects, built for AI agent memory and realtime state. Here's what its cold-start and cost numbers actually are, what self-hosting it takes, and whether a container-only PaaS like bex needs to build its own actor primitive.
Sealos and the Rise of Prompt-to-Deploy AI-Native Clouds: What 'Deploy Anything With a Prompt' Gets Right That a Git-Push PaaS Still Doesn't
Sealos pitches 'deploy anything with a prompt' as a new AI-native category, but its own architecture shows agent skills calling the same Kubernetes CRDs the dashboard already used. Here's the actual dividing line between a real AI-native platform and a chat UI bolted onto an existing API.
Broadcom Donated Velero to CNCF — Here's the Backup Checklist a Self-Hosted PaaS Should Actually Run
Broadcom handed Velero to the CNCF Sandbox, but Sandbox-tier projects still archive at a 15-20% rate within three years. Here's the honest risk read and the concrete backup checklist for running Velero against owned object storage.
Run Your Own Cloudflare Workers: What OpenWorkers and Vorker Actually Buy You Over a Container-Based PaaS
OpenWorkers and Vorker let you run Cloudflare Workers' V8-isolate model on your own hardware. Here's the real cold-start, memory, and filesystem tradeoff against a container-based PaaS, with numbers.
Agent-Substrate vs. Kubernetes' Agent-Sandbox CRD: What 30x Pod Oversubscription Costs You in Scheduling Complexity
Agent-Substrate claims 30x pod oversubscription over Kubernetes' Agent-Sandbox CRD by sharing a worker pool instead of pinning one pod per agent. Here's exactly what that density costs in scheduling complexity, and when it's worth paying for.
Anthropic Acquires Bun to Power Claude Code's $1B Run-Rate: What an AI-Lab-Owned, Still-MIT-Licensed JS Runtime Means for a git-push PaaS's Buildpack Neutrality
Anthropic now owns the runtime Claude Code ships as an executable on. Seven months of release history, a $165K AI-driven Zig-to-Rust rewrite, and what a self-hosted PaaS's build layer should actually do about the risk.
MCP Gives Agents Tools. Agent Skills Gives Them Runbooks.
A folder with a SKILL.md file can hold an entire deploy-troubleshooting runbook and cost an agent almost nothing until it's actually needed. Here's the token math on why that beats stuffing the same runbook into an MCP tool description — and what a deploy-from-chat MCP server should borrow from it.
Building a Bex Golden Path in Backstage: A Working Software Template, Not a Wiki Page
A hands-on walkthrough of building a real Backstage Software Template — template.yaml, a custom scaffolder action, and the generated manifest — that scaffolds a service straight onto a self-hosted PaaS fleet, no wiki page required.