Flisk
Your AI agent builds the app. Flisk runs it on your laptop.
26 followers
Your AI agent builds the app. Flisk runs it on your laptop.
26 followers
Ask Claude Code, Cursor or Codex for an app with a website, an API, a database and an AI agent. Flisk gets every piece running together on your laptop with one command. Change the code and the app updates instantly. When something breaks, your agent can see what went wrong and fix it. Big project? Start just the part you are working on. Free, open-source core that runs entirely on your machine. Already used by Lekast, MakeOffer, mfin.services and Red Moon Digital.
Interactive
Free Options
Launch Team / Built With
Render Sandboxes: a computer for every agentIsolated compute, persistent state and a real filesystem
Promoted
Hi Product Hunt 👋
Most apps aren’t one program. They’re a website, a back end and a database that all have to start together.
Getting that running on your own laptop usually means a pile of setup files nobody wants to own. AI helpers make it worse: they write the app just fine, then break the setup and burn your tokens fixing it.
Flisk does that part for you. You (or your AI helper) describe each part of the app in one small file and run one command. The whole app then runs on your computer: open the link it prints and click around. Change something, and it updates as you save.
Try it in one line:
npx @flisk-sh/flisk demo
(The demo needs Docker. Smaller apps run without it.)
👶 Never written code? There’s a step-by-step guide that starts from “open the Terminal”:
https://github.com/flisk-sh/flisk/blob/main/docs/start/first-time.md
🤔 Not sure it fits your team? Three quick questions, and you get a score:
https://github.com/flisk-sh/flisk#is-flisk-right-for-your-team
✅ What it is: the setup part of building an app. It starts every part, connects them, and keeps them in sync while you work. flisk doctor tells you in plain words what’s missing on your machine.
❌ What it isn’t: hosting. It doesn’t put your app on the internet. And if Docker Compose or Tilt already works for you, keep it.
Free and open source (MIT), no account or key needed. Works on Mac and Linux; on Windows through WSL2 (Ubuntu).
For developers: one service.json per service. Flisk writes the Dockerfiles, the Compose file and the Tiltfile; Docker runs the containers and Tilt live-updates them. flisk up shop starts just that stack, flisk config verify catches drift, and flisk ui is the terminal dashboard. Production stays on Helm, Argo CD or Kustomize.
A question for you: if your app already runs fine on your laptop, would one small file per part be worth it, or just one more thing to keep in sync?
⭐ GitHub: https://github.com/flisk-sh/flisk
🗞 Also on Hacker News today: https://news.ycombinator.com/item?id=50187746
@flisk @privacy_policy Also a maker on this.
Oleksandr covered the what and the why, so I will not repeat it.
If your laptop setup already works, one service.json per part is only worth it when that setup is the thing nobody wants to own. That is the case we kept hitting: the app was fine, the files around it were not, and an AI helper made the files worse.
npx @flisk-sh/flisk demo is the fastest way to see if that is your problem too.
@flisk @kburym also best way to try is to use claude skills for flisk
I've hit exactly the failure mode you describe — the app code is fine, and then an agent rewrites a Dockerfile or Tiltfile and I spend the evening untangling setup I never asked it to touch. The command-only service is what caught my eye, since a lot of the local pieces I run don't need Docker at all and shouldn't get one invented for them. If one service.json is generating 17 files, how does starting just one stack work when the other services are down — do their gateway URLs still resolve, or do you stub them until they're started?
@jcbuilds Thanks. That evening is exactly what Flisk is for.
No stubs. The rules are:
In dependsOn: flisk up shop starts the stack plus what it depends on, across stacks too, plus Postgres and other infra. Each service waits until its dependencies are healthy. The URL is the real service.
Not in dependsOn: it stays off. No container means no route, so you get a loud 404, never a fake answer. flisk doctor names the service and URL.
Sablier (paid): mark a service deferStart. Its image builds, but no container starts. The route sits in front of a wake gateway, and the first request starts the service (and its deferred dependencies), then passes the request through. Idle containers stop again after 10 minutes. You can see every route without starting everything.
Command-only: if every service has a command, they run as local processes. No Docker, no Tiltfile, nothing generated. You still get dependency order, health checks, restarts and logs.
One honest limit: a project can’t mix command services and container services yet. flisk up refuses and lists both groups. That’s next on our list.
What does your setup look like?
Its nice, i have contributed to this https://github.com/flisk-sh/flisk/pull/1904#issuecomment-6442540426 as well!
@rahul05ranjan Thanks Rahul! 🙌 Your MCP example in #1904 is exactly what people need to see that Flisk can run an AI agent’s tools next to the app. Thanks for helping make it clearer.
we are #1 repo of the day https://trendshift.io/repositories/266432#achievements
We built Flisk because coding agents are good at app logic and bad at local infrastructure. They rewrite Dockerfiles, break Tiltfiles, and burn tokens fixing the same plumbing.
Flisk is one service.json per service, then flisk up. It writes the Dockerfile, Traefik routes, env, and Tiltfile. Docker runs the containers. Tilt live-updates them. No local Kubernetes, and no Compose file to maintain. Production stays on Helm. Core is MIT. Premium is optional at $19 per developer per month.
A service that only needs a command can run with no Docker. Databases, one gateway URL, and the container you deploy use Docker and Tilt. flisk doctor checks the machine. flisk config verify catches drift if someone edits a generated file by hand.
If Compose or Tilt already works for you, keep it. Flisk is for several services, agents, and MCP servers that should start as one stack.
Repo: github.com/flisk-sh/flisk