superalp1985/fame-knowledge-agent-gateway

FAME Knowledge Agent Gateway is an external memory and tool control system meant to sit alongside AI coding agents. Instead of stuffing everything an agent knows into its active conversation, it keeps a separate knowledge graph, project memory, and a set of rules that govern which tools the agent is allowed to call, all running outside the agent itself with a visual workbench to inspect it. The README is explicit that this is not a typical retrieval system and not simple chat memory. Its stated purpose is to help an agent first align on a goal, then reason through a structured knowledge path rather than a flat search index, then have any tool calls that change something pass through an approval step before they actually run, and finally write successes, failures, and lessons learned back into external memory instead of just letting that information disappear once the conversation ends. The project is organized around a few pieces: a graph based workbench built with React and Three.js for browsing and editing the knowledge net, a runtime server that exposes an HTTP and MCP gateway backed by SQLite, a project memory layer meant to hold engineering context without polluting the core knowledge base, and lightweight indexes for images and other non-text assets so the raw files stay in separate storage rather than the graph itself. Documentation and the codebase are presented in both Chinese and English, with the Chinese material described as the more complete open edition. It ships with npm scripts to install, generate indexes, run the workbench and gateway locally, run tests, and package a portable release, plus a Docker Compose setup. It is released under the Apache License 2.0.