ling-kong-ran/pisper

Pisper – Multi‑Agent Chat & Workflow Platform

What it is – Pisper is a cross‑platform (desktop, terminal, Android/iOS) application that lets you run many AI agents side‑by‑side. It treats each turn of a conversation like a Git commit: you can branch from any completed turn, keep the original thread unchanged, and later merge or explore parallel branches. Each branch runs in its own sandbox with its own model, working directory, and permission set.

Why it matters – By giving agents their own isolated context and the ability to fork conversations, Pisper turns a single chat into a full‑blown productivity workspace. It is built on the open‑source Pi Coding Agent runtime, so it can call large‑language‑model providers (OpenAI, Anthropic, etc.) and execute tools (shell commands, plugins, custom “skills”) without flooding the model prompt with irrelevant data.

Key capabilities

• Branch‑able dialogs – create a new session from any finished turn, keep the original intact, and jump between branches via a visual tree view.
• True parallelism – multiple sessions run concurrently, each with its own model, file system, and permission level; drag‑and‑drop tags to view four panels at once.
• Tool layering – core tools stay resident in the context; additional plugins, MCP services, and skills are loaded on‑demand via a discover/call gateway and discarded after use, keeping prompts short and cache‑friendly.
• Automatic plugin generation – when a capability is missing, Pisper can write, validate and install a local plugin so the next turn can use it.
• Unified UI – a React‑based desktop client, a Ratatui TUI, and native Android/iOS apps share the same Node.js runtime and configuration.
• Secure‑by‑default data handling – the runtime listens only on 127.0.0.1, masks common credential patterns before persisting, and requires explicit user approval before any memory is recalled.
• Remote linking – a phone can run its own local runtime or connect to a desktop over LAN (TLS‑protected) or fallback Iroh P2P; connections are authenticated with device bearer tokens.

How to get started

1. Desktop (recommended) – download the latest installer from the GitHub Releases page. The package includes the UI, TUI, and the Node runtime; no separate Node.js installation is needed.
2. Mobile – install the signed Android APK or the unsigned iOS IPA (requires side‑loading). The app bundles the same runtime and can operate completely offline or pair with a desktop.
3. npm – npm i -g pisper && pisper web launches the web UI and config UI. After the first run, add a model provider and API key via the /provider page.
4. From source – clone the repo, run npm install, then npm run dev. Use the provided scripts to build the desktop WebView (npm run desktop:webview:dev) or the iOS app (npm run build:ios).

Data safety – Pisper does not run any cloud service of its own. All conversation history, plugin code, and configuration live in ~/.pisper/agent (or a directory set by PISPER_AGENT_DIR). Sensitive strings (API keys, JWTs, private keys) are redacted before being written to disk or displayed in the UI. Network access is limited to the configured LLM provider; remote desktop access must be explicitly enabled and is protected by TLS fingerprint verification and per‑device bearer tokens.

License – MIT License.

---

All details above are taken directly from the repository’s README.