Antigravity Terminal Agent Harness

(Powered by FastMCP, Ollama, and Redis)

Architecture Overview

This project separates the AI Interface from the Capabilities Layer through an MCP (Message Communication Protocol) bridge.

1. The Brain (Ollama + Client Scripts): agent_cli.py and discord_bot.py handle user interaction. They maintain conversation state, dynamic system prompts, and automatically detect which tools should be invoked via standard ollama function calling.
2. The Hands (MCP Server): server/mcp_server.py maintains isolated capabilities. If an agent wants to hit the Gematria cache or retrieve long-term memory, it calls the MCP Server over custom standard I/O streams. This ensures core tools never crash the LLM loop and keeps concerns cleanly separated.

Core Modules

• config.json: Master configuration. Holds hyper-parameters (e.g. temperature, max_tool_turns) and the core system_prompt text.
• tools/gematria.py: A massively efficient cipher calculator that analyzes texts, converts words into 9 distinct cryptographic codes, and caches the results directly into a Redis Vector/Hash Database.
• tools/memory.py: Semantic memory engine. When the agent wants to remember something, it uses the Ollama nomic-embed-text model to embed the text into a Vector Field natively sorted in Redis Stack.
• client/agent_cli.py: Your global terminal orchestration script. (Alias: agent).

Setup & Execution

1. Start Redis Backend The semantic database and Gematria cipher caching operate entirely in-memory using Redis Stack.

# We spin this up natively in Docker
docker run -d --name redis-stack-server -p 6379:6379 redis/redis-stack-server:latest

2. The Terminal Agent You can invoke the agent interactively from anywhere in your console:

agent

It will load your terminal interface, hook into the MCP server, load the tools dynamically, and immediately start answering your inputs while having full read/write access to your system tools.

3. The Discord AI Interface Your bot will intercept commands from Discord perfectly synced to the same MCP Server. To run it infinitely:

~/mcp_tools/venv/bin/python ~/mcp_tools/client/discord_bot.py

(Requires the token to be set correctly inside ~/mcp_tools/.env)

mcp_tools