th0th
A self-hosted semantic search and memory layer for AI coding assistants that indexes a codebase with local embeddings, aiming to cut context tokens by around 98 percent, installable via Docker or from source and integrated through an MCP client.
th0th is a semantic search and context-compression system for AI coding assistants, aimed at reducing how much context an assistant needs to load to answer questions about a codebase. Its author reports a roughly 98 percent reduction in context token usage after adopting it in their own workflow.
It runs 100 percent offline by default: a setup script installs and starts Ollama, pulls the bge-m3 embedding model, and creates a local configuration, after which a diagnostic command checks Ollama connectivity, database access, embedding generation and migration status. It exposes an HTTP API and health check on a local port.
It installs through a one-line script in Docker, Docker-build or source modes, and integrates with agent tools such as OpenCode through an MCP package, a plugin, or a locally run MCP client, as well as VS Code and Antigravity through a workspace MCP configuration file.