meme

Long-term memory for AI agents.

meme is a production-grade memory pipeline written in Rust. It persists knowledge extracted from conversations (or raw facts) to disk via LanceDB and tracks every change in a SQLite audit log.

Pipeline

1. Semantic Structured Compression — dialogues are windowed and sent to an LLM which extracts atomic, self-contained [Memory] entries with structured metadata (timestamps, locations, persons, keywords, …).
2. Lifecycle Reconciliation — each new entry is compared against existing memories in a single LLM call that decides ADD / UPDATE / DELETE / NOOP, preventing duplicates and resolving contradictions.
3. Intent-Aware Hybrid Retrieval — queries are analyzed by an LLM to produce a retrieval plan that drives parallel semantic (ANN), lexical (FTS), and structured-metadata searches. Optional reflection rounds iteratively refine coverage.

Quick Start

use meme::{Dialogue, Meme};

# async fn example() -> meme::error::Result<()> {
let meme = Meme::builder()
    .api_key("sk-...")
    .model("gpt-4.1-mini")
    .build()
    .await?;

// Ingest a conversation
meme.add(&[
    Dialogue::new("Alice", "Let's meet at 2pm tomorrow"),
    Dialogue::new("Bob", "Sure, see you at Shibuya station"),
]).await?;
meme.flush().await?;

// Store a fact directly (bypasses dialogue windowing)
meme.put("Alice prefers coffee over tea").await?;

// Hybrid search & Q&A
let memories = meme.search("Alice meeting").await?;
let answer  = meme.ask("When will Alice meet?").await?;

// CRUD
let all = meme.list().await?;
meme.update(all[0].id, "corrected content").await?;
meme.delete(all[0].id).await?;

// Audit trail
let events = meme.history(all[0].id).await?;
# Ok(())
# }

Public API

Crate Layout