Proma

MCP.Pizza Chef: proma-ai

Proma runs on your own computer and lets you chat with several AI models side by side, hand off multi-step projects to an AI agent that can read, write, and organize files on your machine, and trigger those same tasks from a group chat app like Feishu using a bot bridge. You bring your own key from a provider such as OpenAI, Anthropic, or Google, plug it into the settings screen, and your conversations, documents, and saved workflows stay stored locally instead of on someone else's server.

Communication
Web/Research
Writing

Use This MCP client To

Chat with multiple AI models side-by-side to compare their answers Have an AI agent sort, rename, and clean up files for me Turn research and notes into a finished written report Trigger an AI task from my phone using a group chat message Search the web from a chat to get up-to-date answers Save a repeatable task as a skill to reuse anytime Keep my conversations and files stored on my own computer

README

Proma

Proma 是一个本地优先的 AI 桌面应用,把多模型 Chat、通用 Agent、工作区、Skills、MCP、远程机器人和记忆能力放在同一个开源客户端里。

它不是只面向闲聊的聊天框,而是一个可以长期沉淀个人工作流的 Agent 工作台:简单问题用 Chat,复杂任务交给 Agent,数据和配置尽量留在本地。

Proma 海报

English README | 新手教程 | 下载开源版 | 下载商业版

最新思考 | 2026 Q2–Q3:勇敢地解决真实的问题 — Proactive · 个人注意力 · 团队协作 | 往期思考:2026 Q1

现在能做什么

  • Chat 模式:多模型对话、附件解析、图片输入、Markdown / Mermaid / KaTeX / 代码高亮、并排对话、系统提示词、上下文管理。
  • Agent 模式:内置 Claude Agent SDK 与 Pi Agent SDK 两套运行时;支持工作区隔离、权限模式、文件操作、长任务流式输出、计划确认和用户追问。Claude 是默认内核,Pi 可在实验性设置中开启。
  • 协作与任务:复杂任务可拆分为可追踪的协作子 Agent / Task,并在消息流中展示调用过程和结果。
  • Skills、MCP 与项目根目录:每个 Proma 项目独立配置 Skills 与 MCP Server。项目文件可使用用户选择的本地项目根目录,也可使用 Proma 托管的空白项目目录;本地项目配置不会被自动导入。
  • 远程机器人:支持飞书 / Lark 机器人桥接,并已提供钉钉、微信桥接入口,用手机或群聊触发本机 Agent 工作流。
  • 记忆与工具:Chat 和 Agent 可共享记忆能力,并支持联网搜索、内置 Chat 工具、Agent 推荐等辅助能力。
  • 本地优先:会话、工作区、附件、配置、Skills 等默认存储在 ~/.proma/,使用 JSON / JSONL 文件组织,不依赖本地数据库。
  • 桌面体验:自动更新、代理设置、文件预览、全局快捷键、快速任务窗口、语音输入、亮色 / 暗色 / 跟随系统主题。

快速开始

下载安装

从 GitHub Releases 下载开源版本,提供 macOS Apple Silicon、macOS Intel 和 Windows 安装包。

开源版可独立使用,并支持自行配置 AI 供应商渠道。如果你更希望使用 Proma 提供的内置模型渠道和订阅方案,也可以按需了解 Proma 商业版。两个版本面向不同的使用偏好,你可以自由选择适合自己的版本。

Proma FAQ

Can I use this to have an AI clean up and organize files on my computer?
Yes — switch Proma into Agent mode and it can read, create, rename, and organize files inside a project folder you choose, showing you what it changed so you can approve it.
Can I use this to write a report from research I've already collected?
Yes — hand your notes or documents to the agent and ask it to draft, summarize, or compare them into a finished write-up you can then edit yourself.
Do I need an account or a key from an AI provider to use this?
Yes. Proma itself is free to download, but you connect it to an AI provider by adding your own key from a service such as OpenAI, Anthropic (Claude), or Google (Gemini) — or you can subscribe to Proma's own hosted plan to skip that step.
How hard is it to set up?
There is some technical work involved: you install a desktop app, then add at least one AI provider key in the settings, and the full Agent features also expect Git and Node.js (or Bun) to already be installed on your computer.
Which AI models can I use inside Proma?
You can connect Claude, GPT-4 and other OpenAI models, Gemini, DeepSeek, and several other providers, and switch between them within the same conversation.
Can Proma send or receive messages in a group chat app?
Yes — you can bridge Proma to a Feishu (Lark) group chat, with DingTalk and WeChat bridges also available, so a message in the group can trigger an AI task on your computer.
Is my data kept private?
Your conversations, files, and settings are stored locally on your computer rather than uploaded to a company's server, and any provider keys you enter are encrypted on your device.
Does this work on both Mac and Windows?
Yes — Proma provides installers for Apple Silicon Macs, Intel Macs, and Windows.
Can I try it without paying for anything?
Yes — the free, open-source version works on its own as long as you supply your own AI provider key; a separate paid plan adds a ready-to-use hosted model option and team features.