⚡ Vibe Engine v11 (Opencode Edition)

What's this?

Vibe Engine v11 is a working local-first agent runtime that:

• Runs large language models via Ollama (Qwen, Mistral, etc.)
• Exposes an OpenAI-compatible API
• Enables tool calling (write, bash, edit, grep, etc.)
• Streams responses in real-time
• Auto-deploys via tunnel for remote access
• Handles unstable model outputs via multi-format parsing fallback

Basically: a self-hosted coding agent backend that actually works.

Try it out:

Notes

• Tunneled using localtunnel (yeah, I know Tailscale / Cloudflare exist — Kaggle hates them)
• This is the last stable version before major experimental changes (v12–v18)

---

Experimental direction (v12–v18)

• Moving from Ollama → vLLM (performance + better tool handling)
• Testing sharper models (e.g. IQuest Coder v1)
• Tried shifting to Claude Code… (didn’t go well..at all)

---

Contribute / Fork

If this looks interesting:

Fork → break it → improve it → push

Do drop a ⭐ if you loved this..Would appreciate

---

Built at 17. Mostly out of boredom. Didn't want to pay for API costs. AI did the heavy lifting. I just kept hitting run until it worked. v18 is where I stopped caring enough to fix it further. v11 is the one that actually works reliably. And here ya go..