Tulimoa Memory

A persistent brain for your agent: experiences by day, consolidation into a knowledge graph overnight, recall by association. And the two ways to give your agent a memory.

Last updated Sep 12, 2026

Tulimoa Memory gives your agent a brain that works like one. During the day it collects experiences: facts, decisions, tool results, notes. At night a librarian model consolidates them into a long-term knowledge graph. Recall works by association, so one cue brings back the whole thread, and every morning there is a briefing of what changed. Your agent remembers across sessions, projects and restarts.

How the brain works#

1. Experiences by day. Every remember, checkpoint, batch write or imported document is stored as an experience the moment it happens. Personal data is replaced by placeholders before it is saved.
2. Sleep at night. Once a night a librarian model reads what the day left behind, merges duplicates, turns details into durable knowledge, links it into the graph and lets unused knowledge fade. Your agent never has to tidy up its own memory.
3. Recall by association. recall searches the knowledge graph by meaning and by keyword, pulls in related knowledge and puts warnings about past trouble first. Until the first night it answers from the raw experiences.
4. A briefing in the morning. The agent gets a digest of what was learned, reinforced and forgotten, plus open problems. When it connects, it already sees a short waking state with its goal, open loops, warnings and balance.

Over time your agent knows more while reading less, because it recalls consolidated knowledge instead of re-reading everything. The details are in Sleep, knowledge and the morning briefing.

Two ways to give your agent a memory#

Tulimoa has memory in two places, and today they are separate: what an agent stores through the gateway does not show up in Tulimoa Memory, and the other way round. Pick the one that fits how your agent runs, or use both side by side.

Get started#