💡 The "Why" The bottleneck in AI today isn't the model—it’s the Context Gap. We have incredibly smart agents like Claude and Gemini, but they are "amnesiacs" regarding your personal workflow. Most tools try to solve this with static system prompts, which are brittle and don't scale.
We built Jumo because we realized that for agents to be truly "coworkers," they need access to a Personal MCP (Model Context Protocol) that is built directly from your lived actions, not just your text instructions.
🏗️ How it Works Jumo doesn't force you into a new chat interface. It lives where you work:
Multimodal Logic Capture: Jumo watches your screen and listens to your intent. It maps your "human" nuances—like why you choose one vendor over another—into a structured logic tree.
Model Agnostic by Design: The output isn't a proprietary file. It’s an MCP Server. This means you can "plug" your context into Claude for one task and Gemini for another. Your memory is portable.
Action-to-Node Synthesis: We built a pipeline that translates visual UI interactions into discrete logical nodes. This turns a 3-minute screen share into a production-ready workflow schema.
🛠️ The Build Multimodal Backbone: Powered by Gemini's vision and audio processing to synchronize "what you do" with "what you say."
MCP Architecture: We implemented the Model Context Protocol as the primary export layer, ensuring that the "memory" Jumo builds is immediately usable by any agentic framework supporting the standard.
Real-time Synthesis: The frontend displays the logic tree as it’s being "born," giving the user a live view of how their actions are being codified.
🧠 Challenges & Learning The biggest hurdle was Agentic Portability. Different models interpret context differently. We had to normalize the extracted "rules" so they remained effective whether the downstream agent was an LLM or a traditional automation engine like N8N.
We learned that Context is the new Code. In a world of agentic workflows, the person who can most effectively package their "how-to" into a machine-readable memory wins.
🚀 What's Next We are moving toward Collaborative Context. Imagine a team where every member contributes to a "Departmental MCP," allowing a new AI agent to be onboarded with the collective intelligence of the entire office in seconds.
Log in or sign up for Devpost to join the conversation.