Back home

Amuse

x+web 0.93
Visit source

Intro

Amuse is the free, open-source personal AI assistant Replit published on 19 November 2026. It runs on your own computer: it can research the web, use a live browser you can watch and take over, set reminders, and build small apps, and it asks before it sends, submits or pays for anything. The official account @Replit posted at 2026-11-19T21:15:44Z (05:15 Beijing time on 20 November). The long text names the product with “Meet Amuse: a free, open-source AI assistant that runs on your own computer,” and links the repo github.com/replit/amuse.

Replit framed the release as a Replit Missions challenge: use its own agent swarm to replicate a leading personal assistant, finish in three days, then run Security Scanner to find and patch vulnerabilities. The repository README describes Amuse as a one-person, one-computer self-hosted app. It needs your own OpenAI API key and PostgreSQL, and does not require a Replit account or another hosted service.

This entry records the product name Amuse. Do not merge it with the already listed orchestration product Replit Missions, with Meta’s Muse family, or with other unrelated apps that also use the name Amuse. The repo was created at 2026-11-19T04:34:09Z (12:34 Beijing time), before this four-hour window; this window records the official @Replit introduction at 05:15 Beijing time.

Advertisement

Explain

Amuse is not a new foundation model. Its pitch is to put a personal assistant on the user’s machine, so memories, chats and login credentials stay local instead of on Replit’s servers. The repo describes naming an agent and opening side chats; web research with sources via OpenAI’s web search; a live browser in the chat that you can watch, take over or stop; reminders, monitors and a background heartbeat that only messages you when something matters; turning goals into step-by-step plans; and documents, PDFs, images, podcasts and small space-like apps. Consequential actions raise approval cards, scoped to once, a task, a site, or always. Logins and API keys sit in an encrypted vault; the model only sees stand-in tokens.

At the time of this scan, the README listed gpt-6.2-sol as the default chat model, gpt-6-luna for quicker steps, and gpt-image-2 for images. Those defaults follow the repo’s .env.example at that moment and may change. There is also an AMUSE_DAILY_LIMIT (2,000 usage units by default) that stops the agent for the day. The project is deliberately narrow: a web app for one person, OpenAI models only, connections through API keys and a live browser rather than MCP, and no phone push notifications yet.

The security model is equally narrow: there is no login, the server listens on 127.0.0.1 only, and it should not be exposed to a network. With Docker Compose, the browser may run without Chromium’s own sandbox, and the container is the isolation boundary. When you evaluate it, separate “Replit used Missions to build this in three days and open-sourced it” from whether the setup, cost and permissions are actually controllable on your machine.

Usage

Start from the official post Meet Amuse and the repo github.com/replit/amuse. The README offers Docker Compose and a from-source path: copy .env.example, set your own OPENAI_API_KEY (and DATABASE_URL if you run from source), follow the current install steps, and open http://localhost:3000. Compose starts PostgreSQL alongside Amuse and publishes the app on 127.0.0.1 only.

It is a fit if you want a personal assistant on your own computer and are willing to use your own OpenAI key. Set a spend limit on the OpenAI project first, and read which actions will ask for approval. Do not expose it to a LAN or the public internet, and do not treat it as a new hosted Replit plan or as a substitute for Replit Missions. Models, limits and install steps follow the repository’s current docs.