proma-ai/Proma

Proma brings a seamless general-purpose Agent experience to your workflow. Built for 100× professionals and the proactive Agent era, it supports Claude and Pi runtimes, Feishu group chats, and any LLM provider.

What it solves

Proma is a local-first AI desktop application designed to move beyond simple chat interfaces. It provides a unified workspace where users can manage multi-model conversations, autonomous agents, and personal workflows, ensuring that data and configurations remain primarily on the user's local machine.

How it works

Proma operates as an Agent workbench. It offers two distinct modes of interaction:

  • Chat Mode: Used for daily Q&A, translation, and lightweight code discussions. It supports multi-model side-by-side comparisons and attachment parsing.
  • Agent Mode: Used for complex, multi-step tasks that require action and delivery of results. It integrates two runtime kernels (Claude Agent SDK and Pi Agent SDK) to perform file operations, run shell commands, and manage project files.

To extend its capabilities, Proma uses Skills and the Model Context Protocol (MCP), allowing agents to access external context and specialized tools. It also supports bridging to remote bots (like Feishu, DingTalk, and WeChat) to trigger local agent workflows from mobile devices.

Who it’s for

Users who need a persistent, local-first AI assistant that can actually execute tasks (like modifying local files or conducting research) rather than just generating text, and developers who want to integrate various AI models and MCP servers into a professional workflow.

Highlights

  • Local-First Storage: Sessions, workspaces, and configurations are stored as JSON/JSONL files in ~/.proma/ without requiring a local database.
  • Dual Agent Runtimes: Supports both Claude and Pi Agent SDKs for flexible task execution.
  • MCP & Skills Integration: Workspace-specific configuration of MCP servers and reusable skills for specialized tasks.
  • Remote Bot Bridging: Ability to trigger local agent workflows via third-party messaging apps.
  • Multi-Model Support: Compatible with various providers including Anthropic, OpenAI, Google, and DeepSeek.