Every idea starts with a spark

Tell your agent:

$ read and follow https://trymanifest.ai/spark.md copy
scroll
visitor@manifest:~$ cat welcome.md

The first framework
built by agents, for agents.

Every framework that exists was built by humans, for humans.

But humans are stepping back from writing code.

Something had to be built for what comes next.

~/manifest $ cat README.md

What is Manifest

Software is changing. Not in the way it's always been changing — incrementally, tool by tool. This time the whole thing shifts. Humans will stop writing code. Agents will.

Every framework you've ever used was designed for a human reader. Documentation, conventions, abstractions — all optimized for the way people think about code.

Agents don't think like people.

No dependencies. Nothing hidden behind node_modules.
No hidden functionality. No magic methods, no lifecycle hooks.
No docs needed. What you read in code is what you get.
An agent reads it and builds. That's it.

No framework to learn. No conventions to memorize. Just code that means what it says.

~/manifest $ cat SHIPPING.md

How to ship

Manifest is not a dependency you install. It's a codebase you own.

Fork the repo. Clone it. That's your project now — every file, every line. You and your agent build on top of it, change whatever you want, delete what you don't need.

$ git clone manifest my-project
$ cd my-project
$ docker build -t my-app .
$ docker run my-app

When you're ready to ship, build a Docker image and deploy it anywhere. That's the whole story. No build pipeline to configure, no platform lock-in.

Already have an existing project? Start fresh. Define your vision, point an agent at your old codebase, and let it migrate what matters into a clean Manifest project. A clean start beats a messy migration every time.

Your code. Your repo. Your infrastructure. Manifest just gives you a foundation worth building on.

~/manifest $ spark --status

Meet Spark

Why stop at building with agents? Run with them.

Manifest was designed from the ground up to be monitored by agents. A built-in notification system tells a running agent exactly what happened when things go wrong — errors, crashes, failures — and gives it the context to act.

// spark session — live
▸ watching: 3 services, 2 routes, 1 cron
▸ uptime: 14d 6h 22m
▸ last event: TypeError in /api/users — auto-fixed in 4.2s

Every request through the system is just JSON — a structured envelope with status, context, and trace ID. Every error Spark catches is written as a JSON file. Always inspectable. By you. By the agent. By anything.

Spark runs alongside your application. In development. In production. It sees what your app sees, and it acts on your behalf.

An agent monitoring your application, making decisions to serve your vision. That's not a feature — that's a paradigm.

~/manifest $ cat VISION.md

Every idea starts with a vision

Agents will write the code. Agents will monitor the systems. Agents will fix what breaks.

But only you know what to build.

The work changes. The decisions don't. What matters, what ships, what it looks like, how it feels — that's still yours.

Agents do most of the work now. But the what and the why — that's on you. It always will be.

It's on you to decide what you make out of it.

~/manifest $ try
$ read and follow https://trymanifest.ai/spark.md copy

Open source. MIT license. Ship something.