Works with your agent

The local runtime your AI agent can start, check and fix.

Run your whole app on your machine, from a Compose file or plain commands. Your agent reads its health and logs, and fixes what breaks.

Watch the tour in your browser

Try it first. No signup, no global install.

npx @flisk-sh/flisk demo
Needs only Node.js 22.12+ No terminal? Take the browser tour

MIT. No telemetry. Releases are signed with Sigstore. Check one yourself

Free, MIT licensed, not a deploy tool. Source

74 developers merged code into Flisk in the last 30 days. 90 since the start.

Teams running flisk up on their own code All adopters

  • 2,710pull requests merged
  • 103releases shipped
  • 33,691npm downloads
  • Last 30 days, all public on GitHub and npm
FliskOSOpen Try it here
All products
Click around, or follow the bug:

This is the real FliskOS interface with sample data. Nothing starts on your machine and nothing leaves your browser. How it was made

Now run your own app.

Pick where you are starting from. Each path is a few commands.

Every service keeps its ports, names and env values. Nothing starts until you run flisk up.

npm install -g @flisk-sh/fliskcd your-projectflisk import .flisk up
How the import works

Prefer pnpm, Bun or a prebuilt binary? See the install options.

Start it. Check it. Fix it.

The same three jobs, from your terminal or from your agent.

Start your whole app from what you already have.

A plain Compose app answered 49 s after flisk import on a clean CI runner. How this was measured

Flisk reads your Compose file. Every service keeps its ports, names and env values.

$ flisk import .
$ flisk up --wait

Check what is running, and why it is not.

In one recorded agent run on a two-service app, status showed both services unhealthy and logs named the cause, DATABASE_URL is not set. Read the run

Doctor checks your machine first and ranks failing checks by what blocks you most.

$ flisk doctor

Fix it without leaving the loop.

In that same run, the agent fixed the code, restarted the stack with down and up, and status showed both services ready. Read the run

Your agent calls these as MCP tools, so it can act on what it reads without you relaying errors.

doctor
up
status
logs
down

# generated [from service.json]

Your local stack, generated from one file.

One real service from the restaurant example. Pick a file, then hover the marked lines to see why each part is there.

service.json is the one file you write per service. Hover the highlighted lines to see what each part feeds.

[ fig. 1 / generated service files ]
reservation-api/service.json json (open the whole file on GitHub)
{ "$schema": "https://flisk.sh/schema.service.json", "schemaVersion": 1,
"appName": "reservation-api", "appType": "backend", "stack": "guest",Identity. Name, kind, and stack. Flisk derives image names, routes, and folder layout from these three values.
"port": 4100, "databaseName": "Flisk_restaurant", "dependsOn": ["restaurant-migrator"], "healthCheckPath": "/health"Runtime contract. Port, database, startup order, and health path. These flow into the Dockerfile, the stack compose file, and Traefik.
}

Numbers you can rerun.

A plain Compose app (a Node API and Postgres) answered 49 s after flisk import on a clean Ubuntu CI runner, images pulled and built from scratch, install not counted. Run · Claim details.

Stability: 1.x local dev. Generated files are a contract; verify with flisk config verify. Core CLI is MIT and needs no key. Premium is optional.

Adopters

Who uses Flisk

Teams that run flisk up on their own code. Listed with their consent in ADOPTERS.md.

5 teams listed

# recently shipped

Recently shipped

Recent additions, available on npm. See every release.

on npm

Simple apps run as plain processes

Give a service a command in its service.json and flisk up starts it as a local process on your machine. All you need is the flisk CLI and your app's own runtime, such as Node.js, Python, Go or Bun. It suits the first apps an AI agent builds: an API, a website, a worker.

{  "appName": "api",  "appType": "backend",  "command": "node server.js",  "port": 4000,  "healthCheckPath": "/health"}

flisk up

What Flisk does not do.

The limits first. Each line links to the file that says it.

  • Not a deploy tool. It runs your app on your machine. Production stays on Helm or whatever you use now. Claims and evidence
  • Plain processes do not reload on save. A service run as a local process needs a stop and start to pick up a change. Container stacks live-update. Quickstart
  • Container images are pinned by version tag, not digest. The golden image pull checks no signature. Security notes
  • The signature covers the checksums, not each binary. There is no SBOM yet. How to verify a release
  • No outside security audit. What was checked, and what was not, is listed line by line. Assurance case
  • Two maintainers, no paid support. Fixes go into the latest release. We aim to reply within one working day. What to expect

Honest answers

The same answers we give in engineering reviews.

We need Kubernetes for production
Local dev doesn't need to match production. The 14 tiny services were healthy in 4.6s on a 16 GB M1 after images existed. Not a cold boot. Production stays on Helm. Claim details.
What about CI/CD?
Flisk is completely decoupled from your CI. Use it for local development and whatever you want for CI/CD. Nothing to migrate away from later.
Is 100 services locally realistic?
As a boot test, yes: this generated /health fixture has 100 small generated services (about 20 lines each) that all became healthy through Traefik in 472 s on a clean CI runner, images built from scratch (the run). It is a scale fixture, not a real ERP product. Real services with real dependencies use more memory. Selective startup runs only what you're working on, and golden L1–L4 layers keep rebuilds near-instant. The discovery daemon regenerates only what changed. Our daily driver is a 14-service stack on a 16 GB M1 MacBook, and it performs superbly; the same engine runs a 27-service monorepo every day.

Priced against what it replaces

Core stays free, and most teams should stay on free. Premium is a separate key for the extras above; no key required to run flisk up. See what waiting costs.

Free
$0 / MIT core

The full CLI, MIT-licensed, with no license key required.

  • flisk up, scaffold, Traefik, Postgres, Tilt live update, golden layers
  • Playwright, C4 and AGENTS.md, if you generate them in your repo
  • No key required to run the core CLI
  • Community support
Start free →

How $17 compares

vs cloud dev envs

Gitpod starts at $20/mo plus compute. Codespaces runs $60–150 per dev per month at full-time usage. Flisk Premium is $17 flat — compute is your machine.

vs enterprise CDEs

Coder and Okteto gate RBAC and SSO behind "contact sales". So does Flisk: SSO is on Enterprise, at custom pricing. Premium is the self-serve plan for the extras listed above.

Get your
Premium license.

Core stays free. Premium is a separate key for the extras above; no key required to run flisk up. Need SSO, air-gapped, or an SLA? Pick Enterprise.

Request a license

Free needs no license key — install and run flisk up. Premium/Enterprise requests are submitted directly — no mail client needed, no spam, unsubscribe anytime.

See it run on your machine.

Try it first. No signup, no global install.

npx @flisk-sh/flisk demo
Needs only Node.js 22.12+ No terminal? Take the browser tour

MIT. No telemetry. Releases are signed with Sigstore. Check one yourself