Now AvailableDedicated AI memory with cryptographic proofs. From $5/mo USD.View pricing →
← All videos
·32:23

Building a TypeScript expert memory as an L2 cache for a fleet of coding agents

A full build: Claude researches TypeScript standards, ingests them into a Parametric Memory substrate, and that substrate then acts as a shared L2 cache so a fleet of agents writes to a consistent expert standard.

About this video

This is the long-form build. Rather than describing the idea, it runs the whole thing end to end: an agent researches current TypeScript standards and practice, that research is ingested into a Parametric Memory substrate, and the substrate is then wired in as a shared L2 cache for a fleet of coding agents.

The problem it addresses is consistency. A fleet of agents working from their own context windows produces a fleet of different house styles, and the drift compounds across a codebase. A shared memory gives every agent the same expert baseline to write against, and the baseline improves as it is corrected rather than being re-derived from scratch each session.

At just over half an hour this is the closest thing to a full implementation walkthrough on the channel — worth watching if you want to see the ingestion and recall steps rather than the summary of them.

What you’ll see

  • Researching a body of standards with an agent, then ingesting the result as durable memory rather than discarding it with the session
  • Wiring a substrate in as a shared L2 cache across multiple agents
  • Why consistency across an agent fleet is a memory problem before it is a prompting problem
  • The full ingestion and recall path, shown rather than summarised

Go deeper