rome-os/rome

Rome – an “agentic OS” for humans and AI agents

What it is – Rome is an open‑source platform that lets you run, persist, and compose AI agents together with custom “apps”. It treats the environment (tools, memory, workflows, UI) as a first‑class artifact that agents can create, modify, and reuse. All capabilities are stored as ordinary source code (git‑tracked) rather than hidden model state, so they survive model changes and can be shared via an app store.

Key ideas

• Guard‑railed, self‑hostable environment – run locally (Docker/Electron) or in Rome Cloud. State lives in Docker volumes or a local DB; telemetry is opt‑in.
• Rome Apps – a package (app.yaml + code) that bundles a purpose‑built UI, persistent data, actions, agents, skills, and hooks. Apps are the noun (persistent product) while workflows are the verb (one‑off tasks).
• Actions & capabilities – typed operations that any agent, routine, or app can invoke. They are version‑controlled and can pause for human approval.
• Model‑agnostic – the platform does not ship its own LLM; you plug in whichever model you prefer.
• Collaboration – humans (guardians) and agents work side‑by‑side; agents can even build new apps for you from plain‑language requests.

How you get started

1. Rome Cloud (preview) – sign‑up at https://romeos.cc/login for a hosted instance.
2. Docker quick‑start – run the provided script:

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

   The dashboard appears at http://localhost:7663.
3. Develop from source – requires Node 24+, pnpm 11.6, Docker‑Compose.

   corepack enable
   pnpm install
   pnpm dev:all   # starts the full stack locally
   

   See CLAUDE.md for the full dev loop.

Typical use cases (as described in the README)

• Code‑review automation – “Fix all P1/P2 review comments until no blockers remain.”
• Email triage – sort, archive, flag, and draft replies automatically.
• Price tracking – monitor a product’s price and notify when it drops below a threshold.
• Customer interviews – conduct semi‑structured interviews, ask follow‑ups, and summarize insights.

Rome Apps – the building block

Component	Role
Actions	Typed functions agents and apps call.
Agents	App‑owned collaborators with their own instructions/tools.
Skills	Plain‑language procedures loaded on demand.
Hooks	Lifecycle extensions (message, event, turn).
Web UI & APIs	Front‑end for the app and HTTP endpoints.
Database & files	Persistent, app‑private state.

Two public SDKs are provided:

• @rome-os/app-runtime – backend side for actions, agents, persistence.
• @rome-os/app-web-sdk – front‑end utilities and a CLI to build and package apps.

Repository layout (monorepo)

• packages/core/ – core runtime, session handling, memory, policies.
• packages/web/ – guardian‑only dashboard UI.
• packages/desktop/ – Electron wrapper for a native desktop client.
• rome_apps/ – first‑party example apps.
• packages/app-runtime-sdk/ & packages/app-web-sdk/ – the public SDKs.
• docs/ – concepts, architecture, vision, and operational docs.

How Rome positions itself

Platform	What persists	How repeated work runs	Primary UI	Hosting
Rome	Actions, skills, apps (git‑tracked) + memory & app data	Saved action invoked directly (no re‑run of LLM)	Purpose‑built apps + chat channels (Telegram, Discord, WhatsApp)	Self‑hosted or Rome Cloud
Grok Bot (xAI)	Per‑bot memory & files	Model runs each time	Chat thread	Hosted only
Hermes Agent (Nous)	Text notes & skill docs	Model or cron script	Chat clients / CLI	Self‑hosted
Manus	Files, tools, DB on a persistent cloud VM	Model or scheduled script	Chat + generated web apps	Hosted (exportable)

Rome’s claim is that by persisting software (actions, apps) rather than just text, the environment compounds: a capability built for one task becomes a reusable building block for future tasks.

Documentation & community

• Product site: https://romeos.cc/
• Docs: https://romeos.cc/docs/rome and https://romeos.cc/docs/building-apps
• App Store: https://romeos.cc/store
• Community: Discord invite (g7EFmEtmqc) and X handle @RomeAILab.

Bottom line – Rome is a genuine, actively maintained open‑source project that provides a full stack for building persistent, composable AI agents and the apps they power. It is aimed at developers and power users who want more than a single chat interaction, enabling long‑running workflows, custom UIs, and exportable code.