Open Source SDK for LLM Memory

Inkess Memory

Memory that thinks like you do.
A plugin-based memory SDK for LLM applications. Remote storage, local runtime, zero latency.

Get Started GitHub

        
import { InkessMemory } from '@inkess/memory'

import { extraction } from '@inkess/plugin-extraction'

import { retrieval } from '@inkess/plugin-retrieval'

const memory = new InkessMemory({

  server: { apiKey: process.env.INKESS_KEY },

  llm: { provider: 'anthropic', apiKey: process.env.ANTHROPIC_KEY },

  plugins: [extraction(), retrieval()],

})

await memory.init() // Pull memories from server

// Wrap any LLM call with automatic memory

const chat = memory.wrap(myLLMCall)

const reply = await chat([{ role: 'user', content: 'Hi!' }])

CORE FEATURES

Built for Developers

Everything you need to give your AI application persistent, intelligent memory.

🧠

Human-like Memory

Four-layer architecture mirrors how humans remember: working, short-term, long-term, and archive. Memories decay, consolidate, and strengthen with use.

🔌

Plugin System

All memory intelligence is pluggable. Use official plugins for extraction, retrieval, decay, consolidation, and conflict detection — or build your own.

⚡

Zero Latency

Git-like architecture: pull memories once, run everything locally. No network round-trip on retrieval. Works offline, syncs when connected.

🔑

One Key, Everywhere

Switch devices, switch projects. Your memories follow you. One API key, one init() call — full context restored.

🤖

Bring Your Own LLM

Use your own API keys for Anthropic, OpenAI, or any OpenAI-compatible provider. Server has zero LLM cost.

📦

MCP Ready

Built-in MCP server adapter. Connect to Claude Desktop, Cursor, or any MCP-compatible client with zero code.

ARCHITECTURE

SDK + Server

SDK runs locally for zero latency. Server handles storage and sync. Like Git for memory.

SDK — Local Runtime

Cache + Plugin Manager + Sync Engine + LLM Provider

Extraction

Retrieval

Decay

Consolidation

Conflict

Your Plugin

↕

Server — Remote Storage

Auth + Sync API + File Storage (R2/S3) + Indexes (KV)

COMPARISON

How We Compare

No vector database. No graph database. No server-side LLM cost.

	Mem0	Zep	Letta	Inkess Memory
Storage	Vector DB	Graph + Vector	PostgreSQL	File Storage
LLM Cost	Server-side	Server-side	Server-side	Client-side (yours)
Latency	Network	Network	Low	Zero
Offline	✗	✗	✓	✓
Memory Model	Flat vector	Temporal graph	OS memory blocks	Human-like layers
Plugin System	✗	✗	✗	✓

USE CASES

From personal assistants to production chatbots

🧠

Long-term Personal Memory

Your AI remembers preferences, habits, and past conversations across sessions. Tell it once, remembered forever.

💻

Context-Aware Coding Assistant

AI remembers your tech stack, coding style, and project decisions. Always follows your conventions.

🎧

Cross-Session Support Bot

Bot remembers previous tickets, solutions, and customer preferences. No repeating yourself.

📦

MCP-Powered Memory

Add persistent memory to Claude Desktop, Cursor, or any MCP client. Zero code — just configure and go.