The Amnesia Problem

Open ChatGPT. Ask it to help with your project. Explain the codebase, the architecture, the conventions, the current sprint goals. Get great help. Close the tab. Open it tomorrow. Start explaining everything again.

This is the default experience with almost every AI tool in 2026. Despite all the advances in reasoning and capability, most agents have the memory of a goldfish.

Why Session-Based Memory Fails

Productivity tools figured this out decades ago. Your IDE remembers your recent files. Your browser remembers your tabs. Your email client remembers your folders and filters. These tools build up context over time because the alternative - reconfiguring everything each session - is unacceptable.

AI agents are the one category of productivity tool that somehow decided amnesia is fine. It's not fine. It means every interaction starts with a context dump, wasting time and losing nuance that's hard to re-explain.

What a Knowledge Graph Changes

A persistent local knowledge graph stores entities and relationships from every session. Your projects, your team members, your preferences, your common workflows - all connected and queryable.

When you say "deploy the app," the agent knows which app, which environment, and which deployment script. When you say "email Sarah about the update," it knows which Sarah and which update. The knowledge graph provides the context that makes natural language actually work naturally.

Building Memory Locally

The key is that this memory stays on your device. It's your personal knowledge graph, not a profile on someone else's server. You own it, you control it, and you can inspect or delete any part of it.

Each session adds to the graph. The agent extracts entities and relationships from your interactions and connects them to existing knowledge. Over time, the graph becomes a rich model of your work life - and the agent becomes genuinely personal.

Fazm is an open source macOS AI agent. Open source on GitHub.