Facts and guesses stay apart
Model inferences start as hypotheses. Accepted cognitions require evidence strength and corroboration.
Portable memory for AI applications
A disciplined long-term memory layer for AI apps: portable across models, traceable to evidence, and careful about the difference between facts and guesses.
For applications that need durable user memory without turning every recorded line into accepted truth.
MemoWeft turns a host application's understanding of a person into durable, reviewable memory: evidence, events, cognitions, confidence, conflicts, recall, and export all stay inside a clean public API.
The core rule is simple: remembering something is not the same as believing it. MemoWeft keeps that boundary explicit in storage and recall.
Model inferences start as hypotheses. Accepted cognitions require evidence strength and corroboration.
Contradictory evidence is surfaced for review instead of being silently overwritten.
Short-lived moods decay while durable preferences can persist and travel across hosts.
The bundled host demonstrates chat with recall, visible memory formation, evidence inspection, cognition cards, and portable bundles. It shows how an app can use Core without reaching into internal stores.
Use MemoWeft through its public entry point. Hosts decide when to ingest evidence, when to organize memory, and how much recall to show.
npm install memoweft
import { createMemoWeftCore } from 'memoweft';
const core = createMemoWeftCore({
dbPath: './memoweft.db'
});
await core.ingestUserMessage({
subjectId: 'user-42',
content: 'I only drink decaf after 3pm.'
});
const turn = await core.handleConversationTurn({
subjectId: 'user-42',
message: 'Recommend an afternoon drink.'
});