Liquid logo

Liquid Agentic Memory Module

Step 1 of 6

Input reaches the chat agent

A multi-turn chat or evaluation query enters the agent. Its system prompt includes a manifest of accessible notebooks, their topics, and token lengths.

The diagram starts with the agent, not retrieval.

No

Write

Read

Small

Large

Raw input stream
multi-turn chat · calendar
user notes · notifications

Chat / background agent
System prompt: accessible notebooks
[name · topic · token length]

Needs
memory?

Answer directly

Append or revise
a narrow Markdown section

Relevant notebook
size?

Read full Markdown file
into answer context

Search indexed sections
with the user query

Compact answer context

Topical Markdown
memory notebooks

File/section metadata
+ FTS index
+ KG / knowledge graph

Async maintenance
merge · split · archive · reindex

Use ← and → keys