Memora

Memora is an open source memory layer for LLM applications that stores facts from conversations as SHA-256 hashed objects on your local machine, inspired by how Git stores code. Every fact you tell an LLM gets extracted, hashed, and committed to a local store. The next conversation, Memora retrieves what is relevant and injects it into the prompt automatically so the LLM always knows what you told it before.

The core problem Memora solves is that every LLM conversation starts from zero. You tell it your name, your project, your deadline, your preferences. Tomorrow you open a new chat and it remembers nothing. Existing solutions like vector databases have no version history, no conflict detection, and no way to verify a stored fact has not been silently changed. Memora fixes this by applying Git's content-addressable architecture to memory instead of code.

Every fact gets a SHA-256 hash as its identity. Every session creates a commit pointing back to the previous one, building a complete chain of memory history. When two facts contradict each other, Memora detects it and resolves it automatically using configurable policies instead of silently overwriting. You can roll back to any past memory state, diff two commits to see what changed, and verify the integrity of every stored fact by recomputing its hash.

The project ships as a Python library, a CLI for inspecting and managing memory from the terminal, and a REST API server. It runs entirely locally with Ollama, meaning zero API keys, zero cloud dependency, and zero data leaving your machine. The demo shows a local chat interface where the LLM remembers facts across completely separate conversations, detects when you contradict yourself, and lets you time-travel back to see what it knew at any previous point in time.