Building Memory v3: Adding Long‑Term Memory to Your Go AI Agent

Introduction

Memory v3 upgrades an AI agent by providing true long‑term memory, allowing the system to remember user preferences, ongoing tasks, and conversation context. Unlike single‑turn Q&A tools, the agent can accumulate experience and act as a collaborative partner.

Memory Interface (MemoryInterface)

type MemoryMessage struct {
    ID      string `json:"id"`
    Content string `json:"content"`
    Type    string `json:"type"`
    Created int64  `json:"created"`
}

type MemoryInterface interface {
    SaveMemory(msg MemoryMessage) error
    SearchMemory(query string, limit int) ([]MemoryMessage, error)
    DeleteMemory(id string) error
    ListMemories(limit int) ([]MemoryMessage, error)
}

The interface supports full‑text search, multiple memory types (preferences, tasks, context), and complete CRUD operations.

SimpleMemoryAdapter Implementation

type SimpleMemoryAdapter struct {
    mu       sync.RWMutex
    memories []MemoryMessage
}

func (m *SimpleMemoryAdapter) SaveMemory(msg MemoryMessage) error { /* ... */ }
func (m *SimpleMemoryAdapter) SearchMemory(query string, limit int) ([]MemoryMessage, error) { /* ... */ }
func (m *SimpleMemoryAdapter) DeleteMemory(id string) error { /* ... */ }
func (m *SimpleMemoryAdapter) ListMemories(limit int) ([]MemoryMessage, error) { /* ... */ }

This in‑memory adapter stores messages for fast access, supports fuzzy queries for easy history lookup, and can later be swapped for SQLite or a vector database.

Using Memory in an Agent

agent := &Agent{
    Memory: NewSimpleMemoryAdapter(),
}

Saving a user preference:

agent.Memory.SaveMemory(MemoryMessage{
    ID:      "001",
    Content: "User prefers fsnotify for file monitoring",
    Type:    "user_pref",
    Created: time.Now().Unix(),
})

Querying recent memories:

mems, _ := agent.Memory.SearchMemory("fsnotify", 5)

The agent can now reference these memories to produce responses that align with the user's habits.

Conclusion and Roadmap

Memory v3 transforms the AI agent from an instant tool into a long‑term collaborator, enabling:

Accumulated experience

Understanding of user preferences

Proactive use of historical information in future answers

Future versions are planned:

Memory v3.5 – vector‑based memory store

Memory v4 – multi‑level memory hierarchy

Memory v5 – reasoning‑chain memory

References

Source code repository:

https://github.com/louis-xie-programmer/easy-agent

Gitee mirror:

https://gitee.com/louis_xie/easy-agent