Crate mem7 

mem7

LLM-powered long-term memory engine with a Rust core and Python bindings.

Quick start

use mem7::{MemoryEngine, MemoryEngineConfig, ChatMessage, MemoryFilter};

#[tokio::main]
async fn main() -> mem7::Result<()> {
    let config = MemoryEngineConfig::default();
    let engine = MemoryEngine::new(config).await?;

    let messages = vec![
        ChatMessage { role: "user".into(), content: "I love Rust".into() },
    ];
    let result = engine.add(&messages, None, None, None).await?;
    println!("{result:?}");

    Ok(())
}

Modules

dedup

Memory deduplication utilities.

embedding

Embedding client trait and the built-in OpenAI-compatible implementation.

history

SQLite-backed audit history.

llm

LLM client trait and the built-in OpenAI-compatible implementation.

vector

Vector index trait and built-in implementations (in-process flat index, Upstash).

Structs

AddResult

Aggregated result of a full add() call.

ChatMessage

A chat message in the conversation.

EmbeddingConfig

Fact

A fact extracted by the LLM from a conversation.

HistoryConfig

LlmConfig

MemoryActionResult

Result of an add() operation for a single memory.

MemoryEngine

The core memory engine. Orchestrates the full add/search/get/update/delete/history pipeline.

MemoryEngineConfig

MemoryEvent

A single entry in the memory audit trail.

MemoryFilter

Filter criteria for querying memories.

MemoryItem

A single memory item stored in the system.

SearchResult

Result of a search() call.

VectorConfig

Enums

Mem7Error

MemoryAction

The type of memory event recorded in the audit trail.

Functions

new_memory_id

Generate a new UUIDv7 memory ID (time-sortable, RFC 9562).

Type Aliases

Result