A local-first AI desktop application that integrates multi-model Chat, general Agent, workspace, Skills, MCP, remote bots, and memory capabilities into a single open-source client.
Proma is a local-first AI desktop application that integrates multi-model Chat, general Agent, workspace, Skills, MCP, remote bots, and memory capabilities into a single open-source client.
It's not just a chatbox for casual conversations, but an Agent workspace designed to accumulate personal workflows over time: use Chat for simple questions, assign complex tasks to the Agent, and keep data and configurations as local as possible.
What you can do now
- Chat mode: Multi-model conversations, attachment parsing, image input, Markdown / Mermaid / KaTeX / code highlighting, side-by-side chats, system prompts, context management.
- Agent mode: General Agent based on @anthropic-ai/claude-agent-sdk, supporting workspace isolation, permission modes, file operations, streaming output for long-running tasks, plan confirmation, and user follow-up.
- SubAgent / Tasks: Complex tasks can be broken down into sub-Agents / Tasks via the Claude Agent SDK tools, with invocation processes and results displayed within the message stream.
- Skills & MCP: Each workspace can independently configure Skills, MCP Server, and workspace files, ideal for accumulating reusable capabilities.
- Remote bots: Supports Feishu / Lark bot bridging, with DingTalk and WeChat bridge entry points already provided, enabling mobile devices or group chats to trigger local Agent workflows.
- Memory and tools: Chat and Agent share memory capabilities, and support web search, built-in Chat tools, Agent recommendations, and other auxiliary features.
- Local-first: Conversations, workspaces, attachments, configurations, Skills, etc., are stored by default in ~/.proma/, organized using JSON / JSONL files, without relying on a local database.
- Desktop experience: Auto-update, proxy settings, file preview, global shortcuts, quick task window, voice input, light / dark / system-following themes.
