Empirical became a set of tools, not just one plugin
The memory layer is getting smaller, clearer, and more useful: one core skill, plus optional tools for work history, hard-earned lessons, and project follow-ups.
There is a difference between giving an AI agent more instructions and giving it a better sense of what matters.
For a while, Empirical treated that problem as one large install. Memory, work history, failure lessons, and follow-ups all lived near each other because they were related. They are related. But they are not the same job.
That distinction matters the first time you install Empirical on a new machine and realize you do not want every possible behavior turned on. Maybe you want the agent to remember project decisions, but not record a work log. Maybe you want a Todo skill for a long-running product project, but you do not need it for a one-off question.
So we split the plugin into a core and optional pieces.
One core memory layer
The core Empirical Memory plugin teaches an agent how to use durable memory: when to look for context, what is worth saving, and what should never be stored.
That is the foundation. It is useful when you open a new session and the agent needs to know what happened before: the decision you already made, the constraint you already explained, or the preference you do not want to repeat.
The rest is optional by design.
Add only the capabilities you need
The Work History plugin is for completed work. It leaves a concise trail so the next agent can understand what changed, why it changed, and what was verified.
The Scar Tissue plugin is for the lessons that should survive the incident. It captures the sharp edge: what surprised us, what caused it, and what to check before walking into the same problem again. The older essay Your AI coding agent needs scar tissue explains why that deserves its own category.
The Todo plugin is for unfinished work that belongs to a real project. It is not a second memory inbox. It is a way to say, “This is the outcome I am trying to reach, and here is the next action.”
A normal project day
Imagine you are building a small product with an AI agent.
On Monday, the core memory skill helps the agent recall that you chose a simple hosted flow instead of adding another dependency. On Tuesday, Work History records that the flow was implemented and tested. On Wednesday, a deployment surprise reveals that a setting you assumed was harmless is actually load-bearing; Scar Tissue records the warning. On Thursday, Todo keeps “write the migration guide” visible as part of the product project.
Four different kinds of continuity. Four different reasons to keep something. One install process that lets you choose.
The plugin repository now documents each package separately: github.com/mattgauzza/empirical-memory-plugin. You can start with the core memory layer and add the optional skills when your workflow earns them.
This is also part of a broader shift away from one giant instruction file. The agent should have native guidance available when it needs it, not a growing wall of global text injected into every conversation. The result is less ceremony and more judgment: remember what matters, record what was learned, and keep unfinished work attached to the project where it belongs.
If the idea of a portable memory layer is new to you, Your AI needs your second brain is the longer explanation. If you are ready to install, start with the plugin installation guide.