Moment v1

tl;dr The best agents in the world are coding agents. They will usher in an era of infinite, instant, completely-personalized software. To fulfill this promise, they will need basic infrastructure for collaboration: authentication, collaborative editing, delivery, and access to private APIs. Moment is a new type of collaborative workspace that makes it easy to build personalized software using AI, with these primitives built-in.

You can skip to the end by watching the demo video, or downloading the app and trying it in under 30 seconds.

Or, you can read on.

Tools for software’s Polaroid era

The promise, at least, is simple: type some words into a box, receive software that does pretty much anything you want. A personal notes app tailored to your specific needs, a calendar manager, an custom NES emulator with memory viz. A child can make an entire app from scratch without any help at all.

But in the context of a team, most of this personalized software does not need to be a complete, totally stand-alone Next.js app, deployed on Kubernetes to the corporate intranet, behind a VPN. Usually it’s just a bit of glue, giving your team a little more context in their day-to-day jobs. It’s a Clickhouse query that powers the customer churn graph, live in your PRD. It’s a button you can click to release your mobile app, right in the release process documentation. It’s a runbook that has live metrics from Datadog, and a table of potentially-affected Kubernetes workloads. It’s a small, internal app that lets you adjust customer orders in the database.

Moment’s vision is to make all personalized software fundamentally collaborative, and useful in the context of a team: built-in, robust, real-time collaborative state; unified search; primitives for interacting with private and third-party APIs (HTTP, gRPC, databases); fine-grained, tunable permissions; all the stuff you want, but don’t want to vibe-code ad hoc.

An environment for building personalized software

Moment is a new type of collaborative workspace that makes it easy to build personalized software using AI.

Most of the time when you use Moment, you’ll write some text, and then ask an agent like Claude Code to add real, live application code. In this video, we use the embedded terminal to ask claude for an NES emulator with a built-in visualization of the entirety of its RAM. It essentially one-shots the task.

Then, with a single click, you can share and collaborate on that document with any other user. You get true, first-class, live collaborative editing. Not a Yjs shim, not something hacked around an existing text editor. High-performance, built-in, live collaborative editing.

Simple though it seems, this demo points at the things that make Moment truly special.

1. Everything is Markdown files

The best agents in the world—Claude Code, Amp, Copilot CLI—are coding agents. They operate on Markdown files, on your actual disk. You can’t use the best agents in the world without having them edit actual files, on actual disk.

We did not cut corners here. We want to use the best agents, so every single Moment document is a collection of plain-old .md files, primarily stored on your disk, in git repositories. Not ejected to .md files. Not synchronized to .md files. Actual .md files, on actual disk.

2. Best-in-class live collaborative editing

So Moment documents are plain-old .md files, stored on your disk, in git repositories. But they are also fundamentally collaboratively-editable. Not kind of collaboratively editable via a Yjs shim that replaces the entire document with every keystroke (incredibly, this is the standard solution right now). Not hacked together to accommodate some existing editor’s limitations. Real, from-scratch, purpose-built collaborative editing:

Always 60fps. Independent of the number of readers or writers (incredibly, this is not true of nearly any editors on the market today).

Unbounded latency. If you lose Internet connectivity, we won’t stop you from editing. (We will warn you that divergent writes might produce odd artifacts though!)

Incremental updates. Yes, as I said before: incredibly, nearly all editors on the market will replace the entire document for every keystroke from a collaborator.

3. Natively programmable

I often ask myself what Notion, Airtable, or Retool would look like if they were invented now. In a lot of ways, the previous generation of SaaS companies feel inherently limited by the pre-AI cost of developing software.

But we are free of that now. We don’t need a built-in table, or even an extensions API. We can generate software to build any widget we want. Including an NES emulator! Or, something a little more serious, like this PR dashboard:

Moment’s secret sauce here is not that it’s just programmable, of course. We have put a lot of work in to make a reactive programming environment that plays well with React. We continue to invest in modern tooling around it, like a built-in Tailwind CSS integration. All the things you need to build whatever your team needs to succeed.

4. Best-in-class support for secrets, private APIs, and authorization

Data is only useful to you if you can consume it. This is why, very early in the journey, we invested a lot of time and energy into building tools to help teams safely and securely make authenticated requests to private and third-party APIs. There are two parts to this solution:

A native, miniature Postman-like interface for building authenticated requests. Ok ok, we liked Yaak’s interface a lot more, so it’s actually similar to that. But if you have used tools like Postman or Yaak, you more or less understand the picture here. You can “template” requests using env vars (e.g., {{env.GITHUB_TOKEN}}), and then the tool takes care of proxying that request with the appropriate credentials.

A native proxy that allows users to make authenticated requests, without disseminating credentials to the client. More abstractly, authors can specify that a request has tokens spliced in opaquely via a proxy. This means that you don’t have to distribute API keys to clients who just want to read data from (say) api.github.com.

In addition to all of these things, we have best-in-class support for Kubernetes in particular. We will soon have best-in-class support for AWS as well.