Cadre-AI is a voice-driven BIM and data assistant that connects to your building model, financial data sources, and the web. It uses the Gemini Live API to interpret natural language queries and commands, then decides which MCP toolsets to invoke (financial queries, web search, or Revit automation) via standard IO. You interact with Cadre-AI through a browser UI or an API client; the system streams microphone audio to the server and returns spoken responses while displaying tool activity and transcripts in real time. This enables hands-free architecture workflows, real-time BIM operations, and up-to-date financial and web information accessed through natural dialogue. To use it, run the server, ensure your Google Gemini API key is set, and connect via the provided UI or client to start talking; the server will route your requests to the appropriate MCP toolsets and return results and spoken replies.

Prerequisites:

• Python 3.11+
• A Gemini API key with Gemini Live API access
• Git

Installation steps:

1. Clone the repository git clone https://github.com/WeberG619/cadre-ai.git cd cadre-ai

2. Install Python dependencies pip install -r requirements.txt

3. Configure environment

   • Copy and edit the environment template if provided cp .env.template .env

     Edit .env to include GOOGLE_API_KEY and other variables

4. Run the server locally python server.py

5. Open the UI (default https://localhost:8443) and start conversing. For Revit integration, ensure you have Revit and the RevitMCPBridge2026 plugin set up as described in the docs. For non-Revit usage, you can disable Revit via REVIT_ENABLED=false in the environment.

Notes:

• If you deploy to Cloud Run or other platforms, configure equivalent environment variables in the deployment settings.

Tips and common issues:

• Ensure GOOGLE_API_KEY is valid and has Gemini API access; without it, the agent cannot process speech or generate tool directives.
• If REVIT_ENABLED is false, Revit-related MCP tools will be disabled; use this for non-Revit environments.
• The server uses MCP over stdio; ensure local tooling (financial_mcp, web_search_mcp, revit_proxy_mcp) is accessible to the main server process.
• PORT can be adjusted; remember to expose the correct port if running behind a reverse proxy or on Cloud Run.
• For browser mic access, generate SSL certs as required by the UI (e.g., self-signed certs) and run via HTTPS.
• If you encounter network or API key errors, verify API key scopes and enable the Gemini Live API in the Google Cloud console.