Crate kronroe_agent_memory 

High-level agent memory API built on Kronroe.

Designed to be a drop-in alternative to Graphiti / mem0 / MemGPT — without the server, without Neo4j, without Python.

Usage

use kronroe_agent_memory::AgentMemory;

let memory = AgentMemory::open("./my-agent.kronroe").unwrap();

// Store a structured fact directly
memory.assert("alice", "works_at", "Acme").unwrap();

// Query everything known about an entity
let facts = memory.facts_about("alice").unwrap();

// Query what we knew at a point in time
let past: kronroe::KronroeTimestamp = "2024-03-01T00:00:00Z".parse().unwrap();
let then = memory.facts_about_at("alice", "works_at", past).unwrap();

Phase 1 API

This crate exposes a practical Phase 1 surface:

• remember(text, episode_id, embedding) — store episodic memory
• recall(query, query_embedding, limit) — retrieve matching facts
• assemble_context(query, query_embedding, max_tokens) — build LLM context

Structs

AgentMemory

High-level agent memory store built on a Kronroe temporal graph.

AssertParams

ConfidenceShift

Confidence movement between two related facts.

FactCorrection

A paired correction event linking the invalidated fact and its replacement.

MemoryHealthReport

Operational memory-quality snapshot for one entity.

RecallForTaskReport

Decision-ready recall result shaped around a concrete user task.

RecallOptions

Options for recall queries, controlling retrieval behaviour.

WhatChangedReport

Summary of changes for one entity since a timestamp.

Enums

ConfidenceFilterMode

Strategy for deciding which confidence signal drives filtering.

Error

RecallScore

Per-channel signal breakdown for a recalled fact.

Functions

is_high_impact_predicate

Type Aliases

Result