AI gets a memory with Obsidian and Claude Code: Here's how it works

 

 

In January 2026, Joshua Pham, a former lead machine learning engineer at Spotify, introduced Enzyme, a persistent memory layer designed to solve "context amnesia" in AI coding agents like Claude Code.

 

The Problem: Context Rot

As of early 2026, users of advanced coding agents have found that while tools like Claude Code can read and write files, they often lose track of logic or specific project rules during long-running sessions. This phenomenon, often called "context amnesia" or "context rot," occurs because standard chat histories are inefficient for storing long-term project state.

 

Enzyme: AI with a Memory

Enzyme serves as an infrastructure solution that gives agents a "memory" for the first time by managing context as a persistent resource rather than ephemeral data:

 

    Persistent Logic: It allows Claude Code to "remember" rules across conversations by reading structured data, such as an Obsidian vault or specialized markdown files, into the agent's active memory. An Obsidian vault is a local folder on your computer that stores your notes as plain text Markdown files, acting as your personal knowledge base where you link ideas, manage projects, and build a "second brain" for capturing fleeting thoughts and detailed literature notes. You can have multiple vaults for different purposes (like work vs. personal) and sync them across devices using services like Dropbox or Obsidian Sync.

    Token Compression: The system provides 2-12x token compression compared to standard keyword searches, allowing agents to load critical context faster and more affordably.

    State Injection: Instead of relying on the AI to recall past dialogue, Enzyme injects verified "state" (e.g., project-specific coding rules) directly into the agent's instructions, ensuring it cannot repeat previously corrected mistakes.

 

Joshua Pham's Shift

Joshua Pham left his role at Spotify (where he led the engineering for AI Playlists and the agentic "DJ") to develop Enzyme independently from his New York apartment. His work represents a broader 2026 trend: the emergence of specialized context engineering tools that sit between the raw AI model and the user's local codebase. Pham founded Enzyme as a proprietary venture. The company focuses on building "generative interfaces" for creative work and has transitioned from an independent apartment-based project into a venture-backed startup.

 

Here is more popular explanation of Enzyme:

 

“Artificial intelligence learns what it has forgotten – and with Enzyme, it gets its own memory for the first time.

 

Claude Code can read and write files. But in larger projects, the agent loses track. A thirty-year-old engineer from New York has built a solution and shows how an infrastructure is emerging around the new AI tools.

 

Thirty years old, alone in a New York apartment, without a steady job for months: Joshua Pham quit his job at Spotify to build a product that addresses a problem that many users won't even realize they have until it's too late. The problem is context – more precisely, too much context."