Our first ship of the VibeKit Launch Week!! Run coding agents on the edge 🔥 We re so excited to partner with the awesome team over at @Cloudflare to ship this!

Introducing @vibekit_sh + Beam 🎉 VibeKit is the safety layer for running Claude Code, Gemini, and other coding agents. You can now run Claude Code, Gemini, and other coding agents in secure cloud sandboxes with: - Snapshotting and forking - Custom Docker images - Ultrafast

👀👀

🚨 New in VibeKit: Git worktree support is here! Now you can manage multiple branches in parallel, without juggling clones. Perfect for feature development, hotfixes and reviews.

Run coding agents in your app with a couple of lines of code.

Big news: VibeKit now supports @modal sandboxes! 🔥 Run interactive code in fully isolated environments, just like teams at Lovable do. npm i @vibe-kit/modal

Superagent makes SOTA models outperform on core safety benchmarks ⚔️ How? It continuously updates its weights in real-time as new threats and adversarial attacks emerge — staying one step ahead instead of playing catch-up. This is a paradigm shift: models that learn safety like

Just shipped some quality-of-life updates to VibeKit. cloneRepository now makes it effortless to manage GitHub projects when building apps inside isolated environments powered by @Cloudflare Sandboxes.

@mattzcarey @zeror404 @Cloudflare I have an old repo, but might recommend @vibekit_sh maintained by @pelaseyed: https://t.co/iEPTdfk3h9 If you're curious to try a direct integration with cloudflare sandbox-sdk (layer on top of containers) - this is a good reference too:

Malicious inputs don’t just break prompts — they can poison your model with hostile data. We built a real-time model that protects your model

Excited to announce AI Firewall — the world’s first firewall purpose-built for AI systems 🥷🔥 Traditional security software was built for traditional systems. AI Firewall helps protect against: 🔺 Prompt injections 🔺 Leaking of sensitive data 🔺 Backdoors

This simple prompt injection got read/write access to 1M repositories. Simple static analysis tools won't spot it. Your LLM observability layer that promises "security" might flag it, but by then the damage is done. In order truly secure model inputs/outputs and actions you

@pelaseyed

thanks for creating @vibekit_sh @pelaseyed

@pelaseyed To some extent it is very funny that we are “standardizing file names” early

Gemma 3 (270M) looks like a big step for compact AI chips. With lower latency and training time/cost you can now basically train models in realtime. Imagine deploying agents to the edge.

You can now use Cursor CLI and Opencode without wrecking anything! ✅ Stops destructive commands from trashing your local machine ✅ Redacts secrets and API keys in real time ✅ Lets you actually see what’s going on npm i -g vibekit

https://t.co/PElSThXEjy

Every developer has their preferred agent - it's a matter of personal taste and preference. But each creates safety blind spots 😵 The fix is simple: Just add 'vibekit' when booting up your favorite coding agent! Same workflow, way safer.

🚀 Excited to announce that VibeKit CLI now supports Cursor CLI and OpenCode! This was by popular demand. Use whichever agent you prefer with the same safety layer.

I attribute this to all of the LLM frameworks out there, in almost all cases they suck.