every app you build, now with its own AI agent.

VibeKit just won LaunchBuff Tournament #6! Sixteen products went into the bracket, builders and founders voted on every matchup, and VibeKit came out on top in all four rounds. That puts us in LaunchBuff's Hall of Fame, alongside every community-voted champion before us. It is a real milestone for us. VibeKit started with a simple idea: anyone should be able to describe an app and have it built, hosted, and looked after by its own AI agent. Seeing a community of builders back that idea, round after round, is a moment worth marking. If you have built with us, shared us, or voted, you helped get us here. Thank you. This is one step on a long road, and there is a lot more coming.

Every app you ship with VibeKit gets its own AI agent that builds it, hosts it, and heals it. From your phone. We're in LaunchBuff Tournament #006 this week. If that sounds useful, a vote here helps: launchbuff.com/products/vibe…

Claude: "API Error: Your computer went to sleep mid-response. The response above may be incomplete." VibeKit: "Task Complete"

VibeKit is now listed on @vibecoding_app, and it is one of the most accurate write-ups of the product anywhere. Persistent agent per app, phone-first workflow, pricing all exactly right. vibecoding.app/tools/vibekit…

"The harder problem was not just getting an agent to write code; that's mostly solved. It was giving that agent the right environment, tools, permissions, integrations, and constraints." This is the entire job, and it does not start at DoorDash scale. Every VibeKit agent runs in its own bubblewrap sandbox, with its own workspace and its own injected credentials, from the first message. Same primitives, one layer down, for people whose whole engineering org is themselves.

VibeKit is @openclaw, hosted. Every app gets a persistent agent, a real GitHub repo, and a live URL. All from your phone.

Where should your OpenClaw agent actually live? Everyone is installing OpenClaw this month, and the question that decides whether it becomes your assistant or stays a weekend demo is not which model you pick. It is where the agent runs. An agent that dies when your laptop lid closes is not an assistant. We run OpenClaw agents at fleet scale (every VibeKit app ships with its own persistent agent), so here is the honest rundown of the five real hosting options, including the failure modes the tutorials skip. 1. Your own machine. Free, running in minutes, and the right first step for everyone. But scheduled tasks and long-running work stop the moment the lid closes, and your auth profiles and state live on that one machine. Fine for trying it, wrong as a permanent home. 2. Docker on a $6 VPS. The canonical DIY answer, and it genuinely works if you enjoy operating servers. What you are signing up for: memory is the real constraint, not CPU (2GB is the honest floor, 4GB if the agent does real builds). OpenClaw ships multiple releases a week, so updates are on you, and auto-updating without a health gate occasionally means waking up to a dead agent. And the state directory IS your agent's memory. If your volume strategy is the default, everything it knows is one docker system prune away from gone. 3. Raspberry Pi. Runs fine on ARM, about $3 a month in power. Two things kill Pi agents in practice: SD cards die under sustained agent writes (use an SSD over USB), and reaching your agent from outside your house means maintaining a tunnel, Tailscale, and dynamic DNS forever. Great weekend project. Real infrastructure it is not. 4. Shared hosting. A lot of people are searching for this setup. The honest answer: most shared hosting cannot run OpenClaw properly. You need a persistent Node process and a writable state dir. Panel-based Node slots get your agent killed by the process reaper between requests. It is not hosted, it is intermittently resuscitated. 5. Managed hosting. The provider owns updates, persistence, sandboxing and uptime. You own the conversation with your agent. Disclosure: this is our category. Every VibeKit app gets a persistent OpenClaw agent that builds the app, hosts it on a live domain, and keeps working after you close your phone. Free tier, BYOK welcome. One fleet lesson that applies to every option: an agent with shell access on a public box is a real attack surface. Sandbox it (we use bwrap in production), and never leave provider keys in world-readable files. The mistake is not picking any particular option. It is picking one whose maintenance burden you will quietly stop paying three weeks in. A half-maintained agent is worse than none, because you stop trusting it exactly when it was starting to be useful. Full comparison with costs: vibekit.bot/blog/openclaw-ho…

Free tier update: you now get two hosted apps instead of one. Too many of you were deleting your first app just to try a second idea. That felt wrong. Build both. Live now on iOS + web, no update needed.

Your app, your model. Claude Fable 5 is now selectable in VibeKit alongside Sonnet 5 and every other model - swap anytime, right from your phone.

A renamed model slug shouldn't be something your users discover for you. We now catch a vanished OpenRouter default before it ever hits a turn. 🛡️

VibeKit just got accepted into the @Apple Small Business Program! Lower App Store fees = more going straight back into the product (and into you shipping apps from your phone). Onward 📈

introducing Switchback — a per-turn cascade model router for AI agents classify each turn → start on the cheapest model in your ladder → escalate mid-turn only when the cheap one fumbles brand-locked. open source. MIT. npm i vibekit-switchback npx vibekit-switchback-mcp

Opus 4.8 NOW LIVE ON VIBEKIT!

tell Claude Desktop: "deploy this repo to vibekit, set OPENAI_API_KEY, then run QA" and it does it. all of it. vibekit-mcp 0.6.0 — 31 MCP tools live now.