Getting Started with Engram

Welcome to Engram's official documentation. Engram is a local-first causal knowledge graph designed specifically for developer decision intelligence.

Unlike generic conversational memory tools that simply store raw strings or vector chunks, Engram intercepts session events, parses the technical workflow via a multi-step LangGraph pipeline, and structurally graphs decisions, rejections, and context.

Engram vs Mem0 / Claude Memory

Why build another memory tool? The AI ecosystem is currently flooded with vector-based RAG memory tools. However, they fundamentally fail for deep engineering workflows.

1. The "Counterfactual" Gap

Standard memory tools (like Mem0) only store what you did. Engram specifically extracts and stores what you rejected and why. If you previously rejected MongoDB for latency issues, Engram surfaces that warning the next time you consider it.

2. Temporal Evolution vs Flat Memory

In most tools, memory gets stale fast. Engram features an active Epistemic Weighting Engine that runs locally in the background. It automatically decays older architectural choices, overwriting or establishing contradictions across projects as your decisions evolve.

3. Causality, not just Similarity

If you search for "database" in a traditional vector database, you get 10 text matches. Engram returns the full causal chain: "You chose PostgreSQL because of the previous auth refactor."

Quickstart Installation

For the sake of privacy and data ownership, Engram runs completely independently on your local machine using Neo4j and local ChromaDB.

Prerequisites

• Python 3.11 or higher.
• Neo4j instance (AuraDB Free Tier or Local Desktop).
• API key for Gemini (Gemini 1.5 Pro via API) or an equivalent LLM.

Installation

Clone the repository and install the daemon.