Engrammic
Guides

Load the EAG Guide

Make your agent proactive with memory

What is the EAG Guide?

The EAG (Epistemic Agent Guide) teaches your agent when to store and recall, not just how.

Without this, your agent waits to be asked. With it, memory is automatic.

By default, your agent has the tools but not the instincts. The EAG guide installs a cognitive pattern that makes your agent proactively recall relevant context, store preferences and corrections without prompting, and explain its reasoning on demand.

Load the Guide

Ask your agent to load the onboarding pattern

Say this to your agent:

Session
>Load the engrammic onboarding pattern

The agent calls patterns with name: onboarding and reads the full cognitive guide.

Verify it loaded

Ask your agent:

Session
>What memory behaviors did you just load?
[agent] I now have proactive memory behavior. I'll: - Recall automatically when topics overlap prior work - Store preferences, corrections, and insights without being asked - Trace my beliefs when you ask "why"

See It Work

Mention a preference

Tell your agent something once:

Session
>I prefer functional components over class components
[agent] Noted. Storing preference... ✓ remember called key: coding-style/react value: functional components preferred Got it. I'll use functional components going forward.

Reference a prior session

Open a fresh session and ask about something from before:

New Session
>What do you know about my React preferences?
[agent] Checking memory... ✓ recall called query: React preferences From a previous session: you prefer functional components over class components.

Ask why it believes something

Ask your agent to explain its reasoning:

Session
>Why do you think we use OAuth2 here?
[agent] Tracing belief... ✓ trace called query: OAuth2 authentication I believe this because in session 2025-05-20 you confirmed the API uses OAuth2 with PKCE. That was stored as a validated fact at the time.

Full Onboarding Example

Here is what loading the guide looks like end-to-end:

Session
>Load the engrammic onboarding pattern
[agent] Loading EAG cognitive guide... ✓ patterns called name: onboarding I now have proactive memory behavior. I'll: - Recall automatically when topics overlap prior work - Store preferences, corrections, and insights without being asked - Trace my beliefs when you ask "why"

Common Patterns

Store Preferences

When you state a preference, the agent stores it:

"I always use pnpm, never npm"

Next session, pnpm install runs automatically.

Store Project Context

For facts that need evidence, use learn:

"This repo uses a monorepo structure with pnpm workspaces."

The difference: remember is for observations, learn is for claims with evidence.

Update Existing Knowledge

To correct or update something:

  1. Recall the existing knowledge first
  2. Store the update with supersedes pointing to the old node

The old knowledge is kept for history but marked as superseded.

Connect concepts for richer recall:

"Link my TypeScript preference to my preference for static typing"

Now when you ask about either, the agent can traverse the link.

What Carries Over

Between sessions:

  • Everything stored via remember, learn, or believe
  • Links between knowledge nodes
  • Evidence chains

What doesn't carry over:

  • The conversation itself (that's your AI tool's context window)
  • Anything you discussed but didn't explicitly store

On this page