Workspace Mind

The Multi-Silo Autonomous Agent for the Modern Enterprise.

Comment

Inspiration

As developers and founders, we spend hours every day in "Information Chaos." We constantly switch tabs to find a deadline in a PDF, check a payment status in a spreadsheet, and then draft an email in Gmail. These are Data Silos. I wanted to build a "Neural Bridge" between these silos—a tool that doesn't just search, but reasons across your entire digital life. My mission with SafeCode Labs has always been "Privacy-by-Design," and Workspace Mind is the manifestation of that vision.

What it does

Workspace Mind is an Autonomous AI Command Center for Google Workspace. It uses Gemini 3 to connect Gmail, Sheets, and Docs into a single intelligent interface. Cross-Document Reasoning: It can identify a client’s request in an email, verify their invoice in a spreadsheet, and correct a deadline based on a PDF scope document—all in one turn. Action Agent: It doesn't just talk; it acts. The agent drafts professional replies based on gathered data and sends them directly through the user’s Gmail account. Multimodal Vision: Leveraging Gemini 3’s vision capabilities, the agent can "see" and analyze images and PDFs stored in Google Drive. Neural Trace: A real-time transparency sidebar that allows users to watch the AI’s step-by-step reasoning and tool-calling process.

How we built it

We architected the system as a Dockerized microservices platform deployed on Google Cloud Run. Frontend: Built with Next.js 15, React, and Framer Motion for a high-performance "Digital Cockpit" experience. Backend: A Node.js Express server acting as an MCP (Model Context Protocol) host, managing secure proxy requests to the Google Workspace APIs (Gmail, Sheets, Drive). Intelligence: Powered by the Gemini 3 family via the official Google SDK. We implemented Parallel Tool Execution and a custom Fuzzy Search Algorithm to handle ambiguous file names and high-latency data fetches. Performance: Implemented multi-level caching and exponential backoff to optimize for Gemini’s rate limits while maintaining a seamless user experience.

Challenges we ran into

Deploying a Next.js 15 Standalone application to a containerized environment presented significant environment variable and runtime configuration challenges. We also had to solve complex OAuth2 scope transitions when moving from a "Read-Only" tool to an "Action-Oriented" agent that can send emails. Debugging CORS issues between decoupled Cloud Run services required a rigorous "Security-First" approach to ensure data flowed safely between the frontend and the MCP backend.

Accomplishments that we're proud of

We are incredibly proud of achieving "Closed-Loop Autonomy." Seeing the agent successfully correct a human error regarding a project deadline by cross-referencing a PDF and then drafting/sending a professional email was a "Magic Moment." Additionally, achieving a production-grade deployment on GCP with a highly responsive, mobile-friendly UI proves that agentic workflows are ready for the modern enterprise.

What we learned

Building Workspace Mind was a masterclass in AI-Augmented Architecture. We learned how to collaborate deeply with AI to accelerate the SDLC, moving from concept to a cloud-deployed system in record time. We gained deep insights into the Model Context Protocol, the intricacies of Google’s OAuth2 ecosystem, and the power of Gemini 3’s reasoning capabilities when applied to unstructured business data.

What's next for Workspace Mind

The roadmap for Workspace Mind includes expanding the SafeCode Labs MCP ecosystem to include connectors for Slack, GitHub, and Jira. We plan to implement Semantic Search (Vector Embeddings) to allow the agent to "remember" long-term context across the workspace. Our ultimate goal is to turn Workspace Mind into a multi-user enterprise platform where teams can orchestrate complex projects through a single, shared intelligence.

Built With

Share this project:

Updates