Skip to main content

​
Overview

a21e maintains persistent memory across sessions. When you correct a response, state a preference, or share project context, a21e remembers it and applies it to future requests automatically.

​
How memory works

  1. Capture — after each interaction, a21e extracts memorable facts (preferences, corrections, context)
  2. Store — memories are embedded and stored in your personal or org scope
  3. Retrieve — on each new request, relevant memories are retrieved via semantic search
  4. Apply — retrieved memories are injected into the prompt as calibration context

​
Memory scopes

ScopeVisible toExample
userOnly you”I prefer Tailwind over styled-components”
orgAll org members”Our API uses camelCase for JSON fields”

​
Viewing your memories

curl https://api.a21e.com/v1/memory \
  -H "Authorization: Bearer YOUR_API_KEY"
Returns your stored memories with their scope, content, and creation date.

​
Adding a memory manually

curl -X POST https://api.a21e.com/v1/memory \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Our database uses UUIDs for all primary keys, never auto-incrementing integers",
    "scope": "org"
  }'

​
Deleting a memory

curl -X DELETE https://api.a21e.com/v1/memory/{memory_id} \
  -H "Authorization: Bearer YOUR_API_KEY"

​
Automatic capture

Memory capture happens automatically after Huddle conversations. After a response, you’ll see a Memory capture section below the composer where you can review and confirm what a21e learned. You can pause automatic memory capture in your workspace preferences.

​
Memory in practice

Without memory (first interaction):
“Write a React component” → Uses class components, JavaScript, CSS modules
After correction (“I use functional components with TypeScript and Tailwind”):
“Write a React component” → Uses functional components, TypeScript, Tailwind CSS
The more you use a21e, the less you need to specify.