Stash:賦予 AI 智能體的持久記憶層 | Stash: A Persistent Memory Layer for AI Agents

終結 AI 的『記憶喪失』,讓智能體在對話間持續進化。 | Ending AI amnesia, letting agents evolve continuously across sessions.

🔎 工具速覽 / AT A GLANCE

CategoryAI Infrastructure / Memory Layer
PricingOpen Source (Apache 2.0)
BestForAI Agents, MCP-compatible workflows, Long-term context management
GitHub Stars⭐ 61

🚀 引言 / Introduction

每個 LLM 每次對話都像是在從零開始。Stash 為您的 AI 智能體提供了一個持久記憶層,使其能夠在不同對話階段中記憶、回溯、整合並學習。您再也不需要重複解釋相同的事情。 | Every LLM starts every conversation from zero. Stash provides your AI agent with a persistent memory layer, enabling it to remember, recall, consolidate, and learn across sessions. No more explaining yourself from scratch.

🛠️ 核心功能 / Key Features

Cognitive Layer Transformation: Converts raw episodes into facts, facts into relationships, and eventually into wisdom.

認知層級轉化:將原始片段轉化為事實,事實轉化為關係,最終演進為智慧。

8-Stage Consolidation Pipeline: Deep processing including fact extraction, causal links, goal tracking, and confidence decay.

八階段整合管線:包含事實提取、因果鏈接、目標追蹤及信心衰減等深度處理。

Broad MCP Compatibility: Works with Claude Desktop, Cursor, Windsurf, Cline, Continue, and other MCP-compatible agents.

廣泛的 MCP 兼容性:支持 Claude Desktop, Cursor, Windsurf, Cline, Continue 等多種工具。

Self-hosted Deployment: Single binary, no cloud required, powered by Postgres + pgvector.

私有化部署:單一二進制文件,無需雲端服務,基於 Postgres + pgvector 儲存。

💡 技術亮點 / Tech Highlights

True Persistent Memory: Breaks the conversation window limit to achieve cross-session knowledge accumulation.

真正的持久化記憶:打破對話窗口限制,實現跨會話的知識累積。

High-Performance Processing: The consolidation pipeline only processes new data since the last run for optimal efficiency.

高效能數據處理:整合管線僅處理自上次運行以來的新數據,確保效能。

Total Data Control: Ensures all sensitive memory data remains local through a self-hosted model.

完全掌控數據:透過自託管模式,確保所有敏感記憶數據留在本地。

📦 快速上手 / Quick Start

Clone the repository: git clone https://github.com/alash3al/stash.git

克隆儲存庫:git clone https://github.com/alash3al/stash.git

Configure environment: Copy .env.example to .env and add your API key and model.

配置環境變數:複製 .env.example 為 .env 並填入 API 金鑰與模型資訊。

One-click start: Run docker compose up to deploy Postgres and the MCP server.

一鍵啟動:執行 docker compose up 即可部署 Postgres 與 MCP 伺服器。

準備好試試 Stash:賦予 AI 智能體的持久記憶層 | Stash: A Persistent Memory Layer for AI Agents 了嗎?

Ready to try Stash:賦予 AI 智能體的持久記憶層 | Stash: A Persistent Memory Layer for AI Agents?

前往 GitHub 頁面 →

留言

這個網誌中的熱門文章

[Security] wpa_supplicant setup

[拆機] Nexus5 更換背蓋、電池

[我的MAC Air] 2012年中,MAC Air SSD升級