# Salchi Salchi is a mobile-optimized web GUI for keeping up with coding-agent sessions from your phone, tablet, or desktop. It currently supports OpenAI/Codex, Claude, Cursor, and OpenCode, with more providers coming soon. Bring the AI subscriptions you already use; Salchi connects to your authenticated provider CLIs instead of reselling tokens.

## Demo https://github.com/user-attachments/assets/c9f8687a-de73-4aef-b614-98a9eb0a5629 ## Why Salchi? Salchi focuses on mobile-first agent consumption: - A mobile-optimized PWA for checking in on coding-agent sessions, reading progress, and responding while away from your main machine. - A web agent GUI you can run from your own VPS or Mac, then access from your phone, tablet, or desktop while keeping the agent runtime on the machine with your projects. That makes Salchi useful when you want: - Use your existing AI subscriptions for Codex, Claude, Cursor, or OpenCode instead of paying for a separate token bundle. - Private remote access through `npx salchi@latest`, the desktop app, or Tailscale Serve without exposing your editor to the public internet. - One web surface for many providers, including Claude, OpenAI/Codex, Cursor, and OpenCode. - Built-in file explorer for browsing project files alongside agent sessions. - Source control views for reviewing local changes without switching tools. - Codex subagents for splitting focused work out from the main conversation. - Independent threads for running separate AI loops while keeping your current session intact. - PDF attachments for asking agents to inspect specs, reports, and other documents. - Codex chat image generation for creating visual assets from the same conversation surface. - PWA push notifications for agent activity. On mobile, install Salchi to the Home Screen first so notifications can work. - Local microphone dictation powered by `whisper.cpp`, with selectable Tiny, Base, and Small English models so you can balance transcription accuracy against CPU and memory use on your VPS or computer. On supported Linux hosts, `whisper.cpp` and the selected model install automatically in the background. - Mobile-first PWA polish for coding-agent workflows that need to stay useful on small screens. ## Installation > [!WARNING] > Salchi currently supports OpenAI/Codex, Claude, Cursor, and OpenCode. > Install and authenticate at least one provider before use: > > - OpenAI/Codex: install [Codex CLI](https://developers.openai.com/codex/cli) and run `codex login` > - Claude: install [Claude Code](https://claude.com/product/claude-code) and run `claude auth login` > - Cursor: install and authenticate the Cursor agent CLI > - OpenCode: install [OpenCode](https://opencode.ai) and run `opencode auth login` ### Run without installing Requires Node.js `^22.16`, `^23.11`, or `>=24.10`. ```bash npx --yes salchi@latest ``` Salchi stays attached to the terminal and prints a pairing URL. Open that URL in your browser. ### Tailscale quick start on macOS Use this when you want to run Salchi on your Mac and access it privately from your phone, tablet, or another computer. This uses Tailscale Serve, not Funnel, so the URL stays inside your tailnet. 1. Create a Tailscale account at [tailscale.com/start](https://tailscale.com/start). This creates your private tailnet. 2. Install [Tailscale](https://tailscale.com/download) on the Mac that will run Salchi and on each device that should open it, then sign in to the same account on all devices. 3. In the [Tailscale DNS settings](https://login.tailscale.com/admin/dns), keep MagicDNS enabled and enable HTTPS certificates. 4. Install and authenticate at least one provider CLI from the warning above on the Mac. 5. Open Terminal on the Mac, change into the project you want Salchi to manage, and confirm Tailscale is connected: ```bash cd ~/projects/my-app tailscale status ``` 6. Start a headless Salchi server and keep macOS awake while it is running: ```bash caffeinate -ims npx salchi@latest serve --tailscale-serve --port 4888 ``` Salchi prints a pairing URL like: ```text https://your-mac.your-tailnet.ts.net/pair#token=... ``` Open that URL from another device signed into the same tailnet. 7. **Optional:** To enable PWA push notifications, follow [Tailscale Serve for PWA Push Notifications](https://github.com/JoseRFelix/salchi/blob/main/docs/tailscale-serve-pwa-push.md). Use a non-default Tailscale HTTPS port with: ```bash caffeinate -ims npx salchi@latest serve \ --tailscale-serve \ --tailscale-serve-port 8443 \ --port 4888 ``` Stop the default Tailscale Serve route afterward with: ```bash tailscale serve --https=443 off ``` If you used `--tailscale-serve-port 8443`, stop that route with `tailscale serve --https=8443 off`. `caffeinate` keeps macOS awake while Salchi is running, but it will not reliably keep a Mac awake with the lid closed. Keep the lid open, or use clamshell mode with power connected and an external display, keyboard, and mouse. ### Desktop app Install the latest version of the desktop app from [GitHub Releases](https://github.com/JoseRFelix/salchi/releases). ## Some notes We are very very early in this project. Expect bugs. We are not accepting contributions yet. Observability guide: [docs/observability.md](https://github.com/JoseRFelix/salchi/blob/main/docs/observability.md) Local dictation guide: [docs/local-dictation.md](https://github.com/JoseRFelix/salchi/blob/main/docs/local-dictation.md) ## Screenshots

## If you REALLY want to contribute still.... read this first Before local development, prepare the environment and install dependencies: ```bash # Optional: only needed if you use mise for dev tool management. mise install pnpm install --frozen-lockfile ``` Common development commands: ```bash pnpm dev pnpm dev:desktop pnpm build pnpm start ``` Repository quality checks use Vite+ directly: ```bash vp fmt --check vp lint --report-unused-disable-directives vp run typecheck vp run test ``` Read [CONTRIBUTING.md](https://github.com/JoseRFelix/salchi/blob/main/CONTRIBUTING.md) before opening an issue or PR.