Reference

Everything Orepo does

Orepo is a mobile-first AI builder with a Monaco-based Code IDE for researchers and power users. You describe a website, a shop, or an app change in chat, edit in a file tree and tabs, review diffs, watch a live preview, and publish a public link. Website projects use a Remote IDE strip: by default the plan-server project folder (`vps/<user>/projects/<id>`), or — when a Remote machine has SSH credentials — that host as the live filesystem over SSH/SFTP. Optional code-server on the host adds a VS Code / Extensions surface (Open VSX). WinSCP-style file/folder transfer and agent write-through use the same FS backend. It also opens an existing GitHub repository (public, or private after you connect GitHub), edits those files in place, and can Run the project's own server or Expo web bundler when an E2B key is available. It runs in the browser and as a phone app (iOS-first, App Store submission is prepared but not listed yet) on one account.

What's different now

Compared with earlier chat-only or Monaco-without-remote builds:

  • Code is a real IDE surface: file tree, Monaco tabs, Diff, optional preview, and chat beside.
  • Remote IDE: plan-server folder by default; with SSH credentials on a tagged Remote machine, Monaco / Upload / terminal / agent write-through use SSH/SFTP under the project root (default `~/orepo-projects/<id>`).
  • Optional VS Code / Extensions: idempotent code-server install on the SSH host, proxied through Orepo (owner-only). Extensions come from Open VSX, not the Microsoft Marketplace. Monaco remains the default fallback.
  • WinSCP-style File transfer (Composer → File transfer, and Upload in Code): upload files and folders, browse the live path, progress, and download.
  • Git Commit, Open PR, and Push remain in the Code IDE chrome where already shipped.

Still not

  • Not a native desktop app.
  • Not the Microsoft VS Code Marketplace (code-server uses Open VSX). Not full Cursor-depth Remote-SSH / multi-root indexing / ghost-text.
  • Not a private dedicated VPS or bare-metal machine that Orepo provisions for you — you bring a host you already have.

Build

  • Chat prompt to a website: the files, the chat, and the preview stay on your account.
  • Brainstorm first: a canvas and a few questions, and nothing is built until the answers are in.
  • Live iframe preview beside the chat for index.html pages, in a phone size or a computer size.
  • Run: syncs the project to the plan server and starts a static server, a Node/Vite-style dev server, or Expo web. Returns a live preview URL when an E2B key is available (your key in Settings → Agents, or the server E2B_API_KEY).
  • Shops: products, prices, and a cart. If you already have a Shopify store, give the .myshopify.com address and the page checkout points at that store.
  • SEO basics: a title and a meta description are written into the page files. No ranking is promised.
  • Design: generate a photo or a short video from a prompt. It spends the same monthly allowance as chat.

Code workspace (Monaco IDE)

  • On desktop, Code opens a real IDE layout: file tree | editor tabs + Monaco | optional preview or side-by-side Diff. Chat stays in a side panel.
  • On phone, Code stays chat-first; open the Code button to focus the Monaco editor panel.
  • Diff toggles before vs proposedContents for agent edits (Accept / Reject on the active file). Accept writes through to the live FS backend (SSH or plan-server).
  • Type @ in a message to point at a file, a folder, the docs, git, the web, or the whole codebase.
  • Rules live in .orepo/rules.md and Orepo follows them. Start a message with remember to save a memory.
  • Skills: add Markdown files under .orepo/skills/; the agent loads them into the system prompt.
  • Remote IDE strip on website projects: Upload always; Sync on plan-server; Reconnect + VS Code when SSH is live. Manual edits and agent apply/accept write through to the same backend.
  • WinSCP-style File transfer: upload files and folders, browse the live path, progress, and download. Owner-only; paths are traversal-protected under the project root.
  • Tools → Terminal (closed by default). A busy chip shows when a command is running.
  • Terminal: on SSH live FS, allowlisted commands run on the remote project root. Otherwise, without E2B, allowlisted commands on the plan-server folder; with E2B, isolated sandbox.
  • Working status shows while the agent is busy. Messages sent while Orepo is working wait in a queue.
  • Monaco is the default editor. Optional code-server (when SSH is up) adds VS Code UI and Open VSX extensions.

Cloud vs VPS / Remote IDE

  • Cloud: managed path — files stay on Orepo. Terminal and Run use the plan-server project folder, or an isolated E2B sandbox when an E2B key is available.
  • VPS / remote: save a Remote machine (host, user, port, optional remote root). Add an SSH key (preferred) or password — stored encrypted server-side. Tag a project: filesystem APIs and terminal use SSH/SFTP under `~/orepo-projects/<id>` (or your root). Without credentials, fallback is the plan-server folder.
  • VS Code / Extensions: ensure code-server on the host (idempotent), health-check, then open via authenticated Next proxy. Open VSX only. Monaco remains if install/proxy fails.
  • This is not a dedicated private VM that Orepo provisions. You bring the host.

Safety and review

  • After an agent edit, open a file in chat for a line diff (added and removed), or use Diff in the Code IDE against proposedContents.
  • With Ask before applying site edits on, choose Deny, Allow, or Allow every time. Deny restores the files from the checkpoint. Allow / accept writes through to the live FS backend (SSH or plan-server).
  • Checkpoints save the files so you can restore them.
  • Git tools (Commit, Open PR, Push to GitHub) appear in the Code IDE chrome and under the chat Git tab. Push opens a real branch and pull request when GitHub is connected.
  • Search finds text across the project files.
  • Verify lists problems in the files.
  • Security audit looks for secrets, injection, cross-site scripting, and auth mistakes.

Plan and manage

  • Tasks board with Backlog, To do, In progress, and Done.
  • Agents panel for running work and approvals.
  • MCP: add HTTP MCP server URLs in Settings → Agents. The agent can call them with mcp_call.
  • Projects list with chats; one account across the web and the phone app.

Existing projects

  • Import a GitHub repo (owner/repo or a link). Without GitHub login, only public repos. With GitHub connected (Connect GitHub in Settings → Models), private repos work too. Lockfiles, dependencies, and secret env files are left out.
  • Open the sample loads a small Expo club app so you can try the flow without your own repo.
  • An Expo app is recognized from app.json or the Expo dependency. Prompts change the screens that are there. The default preview still uses index.html when present. Press Run to start the Expo web bundler or the project server when an E2B key is available (BYOK or server).
  • Supabase: paste the project URL and the anon key. Tables are read live and rows are not copied. The service role key is refused.
  • Remote machine app: host, user, port, optional remote root, and SSH credentials (key preferred). With credentials, tagged projects use SSH as the live FS; optional code-server for Extensions (Open VSX). Without credentials, Monaco stays on the plan-server folder. Not a dedicated private VPS Orepo hosts for you.

Models

  • Picker: GPT-4o mini (default), GPT-4o, GPT-6 Luna, GPT-6.1 Sol, GPT-6 Astra, Claude Haiku, Claude Fable 5.1, Gemini 2.5 Flash, Gemini 2.5 Pro, and a free local model.
  • Auto picks a model for each message: small edits stay on a cheaper model, a first build or a hard job uses a stronger one.
  • With Intelligence on, a hard job goes to GPT-6 Astra, the flagship, which spends the allowance fastest. Auto never picks Claude Haiku or Claude Fable 5.1.
  • Bring your own key: save OpenAI, Anthropic, or Google keys in Settings → Models. Your keys are tried before the server keys. Optionally save an E2B API key in Settings → Agents for isolated terminal and Run / Expo.
  • You can always choose the model yourself.

Publish

  • Publish makes a public Orepo link. Update link pushes new changes.
  • Download the project as a zip.
  • Connect your own domain.
  • Chats stay private. Publish is for a website project.

Account and settings

  • Email and password account. Sign out and delete the account from Settings. Deletion removes projects, files, chats, and published sites.
  • Settings: Account, General, Plan and usage, Appearance, Agents (sandbox preference, optional E2B key, MCP, skills note), Models (GitHub connect, model BYOK), Notifications, Privacy, Keyboard.
  • Connect GitHub in Settings → Models for private imports, push, and pull requests.
  • The site and the app are in English, French, and Spanish.

Limits

What Orepo does not do today.

  • The Code surface is Monaco by default (file tree, tabs, Diff, preview, Remote IDE strip). Optional code-server on an SSH host adds VS Code UI and Open VSX extensions — not the Microsoft Marketplace, not VS Code settings sync, and not Copilot-style / Cursor-depth ghost-text.
  • SSH live FS requires credentials you store (encrypted at rest). Auto-install of code-server on arbitrary hosts can fail (missing curl, permissions, OS); Monaco remains the fallback. Local/dev can use host `mock.local` or `OREPO_SSH_MOCK=1`.
  • No multi-root Cursor-depth codebase indexing.
  • File APIs are owner-only and path-traversal protected under the plan-server root or the SSH project root.
  • Run and Expo web need an E2B sandbox (your E2B key in Settings → Agents, or server E2B_API_KEY). Without it, the iframe still shows index.html when present, but there is no live bundler URL. The free local terminal is allowlisted and cannot run arbitrary shell.
  • GitHub Connect needs GITHUB_CLIENT_ID and GITHUB_CLIENT_SECRET. Without them, only public repo import works.
  • Push and GitHub PRs need a connected GitHub account and a project imported from owner/repo. Up to 80 files per push.
  • MCP needs reachable HTTP MCP servers you configure. Skills need files you add under .orepo/skills/.
  • Gemini agent turns do not yet run the full tool loop; they return a text reply. OpenAI and Anthropic keep the tool loop.
  • Publishing hosts pages that Orepo builds. It does not host an arbitrary long-running app server for the public link.
  • It does not log in to your Shopify admin and does not edit a live WordPress site.
  • SEO is a title and a meta description in the files. Orepo does not promise Google ranking.
  • The phone app is not in the App Store. A legal company and Apple Developer enrollment are not created by this software.
  • The Free plan has the local model only. Hosted models need a paid plan (or your own BYOK keys on a paid plan).

Plans

Free
€0 · Local model only, no paid APIs
Plus
€25 per month · €23 of models
Pro
€65 per month · €61 of models
Studio
€125 per month · €118 of models

Compared with other tools

Short descriptions from public product pages as of October 2026. Products change, so check each site for current details.

Cursor

A desktop code editor built on VS Code, with agents, completions, and extensions, aimed at working developers.

Strong at: Deep work in a large local codebase: full VS Code editor, extension marketplace, tab completion, your own local terminal and toolchain.

Orepo: Same class of tools for many workflows: agent chat, Monaco IDE (file tree, tabs, Diff), Remote IDE with plan-server or SSH live FS + WinSCP-style transfer, optional code-server (Open VSX) on your host, checkpoints, verify, security, terminal, Run, GitHub OAuth/push/PR, skills, MCP, BYOK models. Orepo extras: publish a public link, phone app, brainstorm canvas, shops, and hosted Run/Expo when E2B is available. Not a local desktop install; not the Microsoft Marketplace.

Replit

A browser-based development environment with hosting and an AI agent.

Strong at: Runs your project's own environment and servers, many languages, deployment of full apps.

Orepo: Orepo can Run static, Node, or Expo web when an E2B key is available. Public publish still hosts pages Orepo builds, not an arbitrary always-on app process.

Lovable

A prompt-to-web-app builder, commonly paired with Supabase.

Strong at: Fast full-stack web app scaffolds with a database from the start.

Orepo: Orepo connects an existing Supabase project by URL and anon key and reads tables live, and it can open an existing GitHub repo. Its phone app uses the same account.

Draftbit

A visual builder for React Native apps.

Strong at: Drag-and-drop native app design with a visual editor.

Orepo: Orepo is chat-first, not a visual canvas builder. It can open an Expo repo, change screens by prompt, and Run Expo web on the plan server.

ChatGPT and Gemini

General AI assistants that can write code and answer questions.

Strong at: Broad knowledge, many tools, and general conversation.

Orepo: Orepo keeps the files, the preview, checkpoints, Run, GitHub, and publishing in one project instead of in a chat transcript. Gemini is also available as a model inside Orepo.

FAQ

What is Orepo?

Orepo is a mobile-first AI builder with a Monaco-based Code IDE for researchers and power users. You describe a website, a shop, or an app change in chat, edit in a file tree and tabs, review diffs, watch a live preview, and publish a public link. Website projects use a Remote IDE strip: by default the plan-server project folder (`vps/<user>/projects/<id>`), or — when a Remote machine has SSH credentials — that host as the live filesystem over SSH/SFTP. Optional code-server on the host adds a VS Code / Extensions surface (Open VSX). WinSCP-style file/folder transfer and agent write-through use the same FS backend. It also opens an existing GitHub repository (public, or private after you connect GitHub), edits those files in place, and can Run the project's own server or Expo web bundler when an E2B key is available. It runs in the browser and as a phone app (iOS-first, App Store submission is prepared but not listed yet) on one account.

Is Orepo an alternative to Cursor?

For many researcher and power-user workflows, yes in the same class of tools: agent, Monaco IDE layout, Remote IDE + file/folder transfer, optional SSH live FS and code-server (Open VSX), diffs, checkpoints, terminal, Run, GitHub push/PR, skills, MCP, BYOK. Cursor remains stronger for local desktop VS Code, Microsoft Marketplace extensions, multi-root indexing, and deep local toolchain work. Orepo adds publish, phone, brainstorm, and shops.

Does Code include a real IDE?

Yes on desktop: file tree, editor tabs, Monaco, Diff against proposed agent edits, Remote IDE strip (Upload; Sync on plan-server; Reconnect / VS Code when SSH is live), Git commit/PR/push, and optional preview. Phone stays chat-first with a focused Code editor panel. Full Microsoft Marketplace extensions are not available; code-server uses Open VSX when SSH is connected.

What is Remote IDE and File transfer?

On website projects, Code talks to a live filesystem: the plan-server folder (`vps/<user>/projects/<id>`) by default, or SSH/SFTP on a tagged Remote machine when credentials are stored. Upload, WinSCP-style transfer, agent accept, and manual edits write through there. Owner-only and path-traversal protected. Optional VS Code opens proxied code-server (Open VSX). Not a dedicated private VPS that Orepo provisions.

Cloud vs VPS — what is the difference?

Cloud means managed Orepo hosting of the project folder (and E2B when keyed). VPS / remote tags a saved machine you already have; with SSH credentials, that host is the live FS (and optional code-server). Without credentials, Remote IDE stays on the plan-server folder.

Can I run Expo or my own dev server?

Yes, with Run, when an E2B key is available (your key in Settings → Agents, or the server E2B_API_KEY). Expo web, a Node/Vite-style start script, or a static file server is started and a live preview URL is shown.

What can the free terminal run?

Without an E2B key, only allowlisted commands inside the project folder on the plan server (for example ls, pwd, cat, git status, scoped npm/node/python). Dangerous or out-of-folder commands are blocked. Add your E2B API key in Settings → Agents for isolated full commands.

Can I open a private GitHub repo?

Yes, after you Connect GitHub in Settings → Models (needs GITHUB_CLIENT_ID and GITHUB_CLIENT_SECRET on the server). Without Connect, only public repos import.

Does Orepo work with Supabase?

Yes. Paste the project URL and the anon key and the tables are read live. Rows are not copied, and the service role key is refused.

Which AI models does Orepo use?

GPT-4o mini (default), GPT-4o, GPT-6 Luna, GPT-6.1 Sol, GPT-6 Astra, Claude Haiku, Claude Fable 5.1, Gemini 2.5 Flash, Gemini 2.5 Pro, and a free local model. Auto picks a model per message. You can bring your own OpenAI, Anthropic, or Google keys.

What are skills and MCP?

Skills are Markdown files in .orepo/skills/ that the agent loads. MCP servers are HTTP endpoints you add in Settings → Agents; the agent calls them with mcp_call.

How much does Orepo cost?

Free has a local model only. Plus is €25 per month with €23 of models, Pro is €65 with €61, and Studio is €125 with €118. Using the model amount does not add a second bill.

Is there a phone app in the App Store?

Not yet. The Expo app and EAS profiles are prepared. You still need an Apple Developer account, App Store Connect, and a legal entity. See /legal.

Can I delete my account?

Yes, in Settings, then Account. Deletion removes projects, files, chats, and published sites.

Which languages are supported?

English, French, and Spanish for the site and the app.