Introduction
Give your agent memory that consolidates what it learns and recalls it when it matters.
Augenta is a memory layer for AI agents. Send it your agent's conversations and documents; Augenta extracts the statements worth keeping, consolidates related ones into memory, and returns the relevant memory when your agent needs it. Your agent reads that memory before it calls its model, so it can act on what was decided, tried or learned before.
The loop is two calls around each model call:
- Before: ask Recall what is relevant to the user's message, and add the result to your model's context.
- After: send the finished turn to Experiences, so later turns and other conversations can recall it.
Add memory to your agent builds that loop step by step. To try the API first without storing anything, the Hello world request sends one message to Consolidate and returns the memory it forms.
Create a key and see what Augenta extracts from one message.
Recall before each model call and add each turn after it.
Load existing documents into a Workspace and recall them.
Retrieve relevant memory for your model.
Read with an AI assistant
Choose Copy for LLM on any page to copy its prose, attributes and examples as Markdown.
View Markdown opens the same content without the website layout. Adding .md to a page
URL also works, and agents can request a page with Accept: text/markdown.
Start with the documentation index to find relevant pages. The full documentation export contains every navigation page and is much larger; individual pages are usually enough for one integration task.
Terms
| Term | Meaning |
|---|---|
| Workspace | Where Augenta stores memory. Recall searches everything in one Workspace. An API key is assigned the Workspaces it may read and write. |
| Experience | One unit of source data you send: a document, or a stretch of a conversation or agent run. |
| Claim | One statement extracted from an experience, such as "Production releases need owner approval". |
| Engram | A group of related claims with a summary: memory consolidated from many experiences. |
| Recall context | The claims and engram summaries Recall returns for a question. Your code adds them to the model's prompt. |