proferai

Profer is a local-first, general-purpose AI Agent desktop application built on the Claude Agent SDK and Pi Agent dual runtime. It is designed to delegate simple questions to Chat, complex tasks to Agent, and layer team collaboration on top of powerful local AI Agent capabilities. **Core Features** - **General Agent**: Based on `@anthropic-ai/claude-agent-sdk`, it supports task graph decomposition, subtask dependency orchestration, streaming output, and plan confirmation, with the ability to switch between Claude / Pi runtimes. - **Collaborative Sub-Agents**: Splits complex tasks in parallel across multiple real sub-sessions that advance independently, then aggregates results with full visibility and traceability throughout. - **Scheduled Task Automation**: Supports persistent scheduling (interval / daily / weekly / monthly), with run history, failure protection, and result review, suitable for unattended scenarios like daily/weekly reports and automated checks. - **Mobile Remote Access**: Connects remotely to the desktop via local HTTP/WS, with mobile support for Chat, settings sync, and auto-reconnect (client in a separate repo, Profer-pocket, based on Capacitor Android). - **Team Workspace**: Invitation-based teams (Owner / Admin / Member / Viewer roles), featuring a Skills sharing marketplace, cloud file sync, and brand customization. - **Multi-Model Chat**: Supports multi-vendor conversations, attachment parsing (PDF / Office / images), Markdown / Mermaid / KaTeX / code highlighting, side-by-side comparison, and context management. - **Skills & MCP**: Each workspace can independently configure Skills and MCP Servers, with a full-screen skills view supporting search, enable, update, import, uninstall, and team publishing. - **Remote Bots**: Supports Feishu / Lark / DingTalk / WeChat bridging, allowing mobile or group chat to trigger local Agent workflows. - **Desktop Experience**: Auto-update, global hotkeys, quick task window, streaming voice input (Doubao), and light / dark / multiple refined themes. **Supported Model Channels** Supports Anthropic, DeepSeek, Kimi API, Kimi Coding Plan, OpenAI, Google, Zhipu AI, MiniMax, Doubao, Tongyi Qianwen, and custom endpoints (OpenAI-compatible). Among these, Anthropic and Anthropic-compatible channels support Agent features, while OpenAI, Google, and custom endpoints support Chat only. **Tech Stack** - Runtime: Bun - Desktop Framework: Electron 43 - Frontend: React 18 + TypeScript + Jotai - Styling: Tailwind CSS + Radix UI - Rich Text / Charts: TipTap · Beautiful Mermaid · KaTeX · Shiki - Build: Vite + esbuild + electron-builder - Agent SDK: `@anthropic-ai/claude-agent-sdk@0.3.201` (Claude) + `@earendil-works/pi-agent@0.82.1` (Pi) dual runtime - Mobile: Capacitor (Android) - Team Backend: Hono + better-sqlite3 + JWT **Quick Start** Download the latest version from GitHub Releases, with installers for macOS Apple Silicon / Intel and Windows. Initial setup requires environment checks (dependencies on Git, Node.js / Bun, and a usable Shell), then add an AI channel in settings and select the default channel, model, and workspace. Team collaboration features require deploying a lightweight backend service (Hono + SQLite), which can be deployed to any Linux server with one click. **Local Development** A Bun workspace monorepo, including packages (shared, core, project-core, session-core, ui), apps (electron, cli), and a server directory. Use `bun install` to install dependencies, `bun run dev` to start development mode, `bun run typecheck` for type checking, and `bun test` to run tests. **Contributing** PRs are welcome, but must follow project conventions: use Bun, use Jotai for state management, disable `any` in TypeScript, and update related types and bridge code when adding new IPC. **License** Profer is developed based on Proma, and the community edition uses the AGPL-3.0 license.