rome-os/rome

A compounding agent OS for recursive agents.

Website · Try Rome Cloud · Documentation · App Store

Most progress in AI comes from scaling models. Rome scales the other axis, the environment: the tools, workflows, memory, and interfaces an agent works within (why this matters).

Rome is a guardrailed environment where human and agent collaborate, and the collaboration compounds. Agents build their own harnesses, design their own SOPs, and orchestrate workflows under your guidance. Proven capabilities stick. Every interaction raises the ceiling for the next.

Rome also works as an open source alternative to Grok Bot and Meta's Muse for persistent agents, scheduled tasks, and custom apps.

Rome Cloud provisions a private Rome environment for each guardian. It is currently available as a preview.

Join the preview →

One script checks for Docker, pulls the published image, and starts Rome:

curl -fsSL https://raw.githubusercontent.com/rome-os/rome/main/scripts/quickstart-docker.sh | bash

Or clone the repository and run the script from it:

git clone https://github.com/rome-os/rome.git
cd rome
./scripts/quickstart-docker.sh

The dashboard comes up at http://localhost:7663, bound to loopback only. First-run onboarding is open to whoever reaches it first, so exposing it beyond the machine takes an explicit --bind. State lives in named Docker volumes, so re-running the script upgrades the container without losing data. Telemetry export stays off unless you set OTEL_EXPORTER_OTLP_ENDPOINT. Run the script with --help for ports, profiles, and the other settings it forwards.

Rome Docker images default to PANTHEON_BASE_ORIGIN=https://romeos.cc. To use another Rome Cloud deployment, pass your origin when starting the container with docker run -e PANTHEON_BASE_ORIGIN=https://cloud.example.com or set it in the Compose .env file. Recreate the container after changing the value. The quickstart script also accepts a shell override:

PANTHEON_BASE_ORIGIN=https://cloud.example.com ./scripts/quickstart-docker.sh

To run this repository from source, you need:

• Node.js 24 or newer
• Corepack and pnpm 11.6
• Docker with Docker Compose

From a checkout of this repository:

corepack enable
pnpm install
pnpm dev:all

pnpm dev:all starts the production-shaped local stack: Rome, observability, routing, and the web development server. It connects to https://romeos.cc by default; set ROME_DEV_PANTHEON_ORIGIN to use another Rome Cloud deployment. The script prints the local URLs and development credentials when startup completes.

This is the contributor development path, not the final production self-hosting distribution. See CLAUDE.md for the complete development loop, container commands, and validation requirements.

Rome App is the new way to interact with your agent.

Chat is a good place to ask for something once. Repeated work deserves a place of its own: an inbox that remembers what was triaged, a code review loop you can inspect, a price tracker that keeps watching, or a morning brief that arrives on schedule.

A Rome App combines a purpose-built interface, agent reasoning, reusable workflows, and persistent data into one installable product. It is not a thin wrapper around a prompt. The app remains useful after the conversation ends, after the browser closes, and when the user comes back tomorrow.

A useful rule of thumb: a workflow is a verb; an app is a noun. Use a workflow to perform a task and return a result. Build an app when the work needs a home of its own, with user-editable data, multiple actions, or a persistent agent.

When there is not an app for what you need, describe it to Rome in plain language. Rome can turn that request into a short specification, scaffold the app or workflow, build it into your instance, and keep iterating with you in the same conversation.

The result is ordinary, git-tracked source code rather than hidden model state. Keep it private, adapt it as your needs change, or publish it for others to install from the App Store. This creates Rome's self-evolution loop:

Describe a need → Rome builds the capability → the app keeps working
       ↑                                              ↓
       └──────────── refine, reuse, and share ────────┘

A Rome App starts with an app.yaml manifest and can ship any mix of:

Together these form a capability: the unit Rome discovers and reuses in later work. The app is that capability's human interface: apps organize human interaction; capabilities organize agent action.

Apps build on two public SDKs:

• @rome-os/app-runtime for backend capabilities.
• @rome-os/app-web-sdk for embedded web interfaces and the Rome build CLI.

Browse the Rome App Store, read the building guide, or start with the app quickstart.

These are the kinds of requests Rome is designed to follow through on:

Rome can handle one-off tasks, scheduled work, long-running follow-through, and purpose-built app experiences without forcing everything into one chat window.

Rome sits where two product waves meet: persistent agents and personal software. The agent products share Rome's thesis that agents should persist and improve. There the comparison is what accumulates, what runs when work repeats, and where you operate the result. The personal software platforms share Rome's thesis that software should be built per person. There the comparison is who and what stands behind the app.

* Hermes and Manus can schedule plain scripts that skip the model. A script answers only to its timer. A Rome action is a building block: agents, apps, and interfaces all call it, and it can pause for approval.

Grok Bot gives persistent bots a shared cloud computer. Its skills and routines support repeated work. Rome adds a runtime you can self-host and apps whose actions, data, and interfaces you can reuse.

Muse handles personal tasks with memory, scheduled work, and interactive artifacts. Rome covers similar workflows through apps, with source code you can modify and a choice of model provider.

Hermes Agent is the closest in spirit: MIT-licensed, self-hostable, with curated memory, self-written skill documents, and a skill marketplace. What persists is text that informs the next reasoning run, plus script-only cron jobs that skip the model but answer only to their timer. Rome persists software: executable actions, app-private databases, and purpose-built interfaces. An action is a building block rather than a loose script: agents, routines, and app interfaces all call the same one, it can pause for approval, and every run is recorded. Rome already discovers and composes capabilities from earlier work, and its agenda extends to consolidating, refactoring, and retiring what goes stale (VISION.md). And Hermes offers one general interface to the agent (chat clients, a desktop app, the terminal), while Rome grows a purpose-built app for each repeated workload.

Manus delegates a goal to an agent on a persistent cloud computer: files stay, installed tools stay, and scheduled jobs, long-running bots, and databases can live on the machine. It also builds full-stack web apps whose code you can export and host anywhere. The difference is what that persistence is made of. A Manus schedule re-runs the agent at metered cost or fires a script parked on the machine, reusable only by finding the file again, while a Rome action is a named operation that every surface can call. A Manus app is a deliverable that stands apart from the agent, while a Rome app is also a capability the agent discovers and reuses in later work. And the Manus agent itself, with its knowledge, playbooks, and machine, runs only in its cloud, while Rome self-hosts.

Wabi approaches from the personal software side: a social platform where anyone prompts a mini app into existence, then shares and remixes it, with Wabi hosting everything. It validates half of Rome's bet, software built per person rather than for the average user. But there is no worker behind the app: a Wabi app is the end product, and you still operate it yourself. A Rome app is the interface to a capability, and the capability faces both ways: you work through its views, and agents work through its actions, skills, and data in later tasks. Every new Wabi app grows your library, and every new Rome capability also extends what the agent can do next. Wabi asks you to describe an app, while Rome asks you to describe an outcome and leaves an app behind when the work deserves one.

The difference is the unit of compounding. Grok Bot compounds hosted bot state and skills, and Muse compounds personal context and artifacts. Hermes compounds the agent's notes, Manus compounds a machine and its files, and Wabi compounds a network of shareable apps. Rome compounds the environment: executable, composable capability owned by you (why that is the durable asset). Building blocks compose: a capability built for one task becomes a part in a more ambitious one, so each request can ask for more than the last. An inventory-aging action and a customer-churn skill, built for separate tasks, later combine to clear seasonal inventory without a broad markdown.

Proven actions close the cost and reliability gap, purpose-built apps close the interface gap, and maintenance keeps the environment coherent as it grows. Because the environment is open, git-tracked, and exportable, it survives a model swap. Users stay for the compounding, not the lock-in.

Rome is a pnpm monorepo.

Common checks:

pnpm typecheck
pnpm test:unit
pnpm lint
pnpm build

• VISION.md — why Rome exists and the product principles it protects.
• docs/concepts/ — canonical domain vocabulary.
• docs/architecture/ — component boundaries and system invariants.
• Using Rome — public guardian documentation.
• Building Rome Apps — public app-author documentation and SDK guides.

Give your agents a place to grow.
Start with one workflow. Rome keeps what works and compounds from there.