s3yed/cognify

Cognify is a memory tool for AI agents that goes beyond simply finding related text. When you hand it documents, it reads them and builds a map of the people, projects, and technologies mentioned, along with how they connect. You can then ask an agent questions like "who owns this project and what does it run on?" and it can trace the relationships to give you a connected answer, not just a list of isolated paragraphs. It works in three steps. First, it breaks your documents into manageable chunks. Second, it sends each chunk to a cheap AI model that pulls out typed facts, like "Sarah works at Acme" or "Acme uses Pathfinder." Third, it saves these facts into a lightweight map (a graph) alongside the original text. When you ask a question, it searches for matching text, then expands outward along the connections it found, gathering related facts so the agent has full context. This is built for developers or founders who want to give an AI agent genuine knowledge of a company's documents, a product's architecture, or customer support history, without setting up a heavy database infrastructure. A startup could feed it their internal wiki and HR handbook, then ask an agent multi-part questions that require connecting information across different documents. What makes this notable is how deliberately lightweight it is. By default, it requires no external database servers and avoids heavy AI frameworks, relying on a small, CPU-only footprint. The entire core is only about a thousand lines of code, meaning a developer can read it in a single sitting. If a project grows, you can scale it up to a more powerful database setup by changing a single setting, without rewriting your code. It also works directly with Claude as a background tool, or supports keeping multiple clients' data safely separated.