CoPaw, a new open source personal AI assistant from Alibaba’s AgentScope team promises long-term memory, flexible local or cloud deployment, and deep integration across messaging platforms—giving developers greater control over how AI assistants operate.
The AgentScope team within the Alibaba ecosystem has released CoPaw, an open-source personal AI assistant designed as a Personal Agent Workstation that users can install, host, and manage themselves. Released under the Apache License 2.0, the platform aims to eliminate vendor lock-in by allowing users to run, modify, and scale the assistant independently.
A key feature of CoPaw is its persistent memory system, powered by ReMe (Remember Me, Refine Me), a specialised memory management framework. Unlike typical stateless chatbots, the system enables the assistant to retain long-term contextual information across interactions. Memory is handled through file-based storage, making it editable and portable, and through a hybrid retrieval mechanism combining semantic vector search (0.7 weight) and BM25 keyword search (0.3 weight). A context compaction feature summarises long conversations to preserve important details without exceeding LLM context limits.
CoPaw supports both local and cloud deployments. It can run entirely offline using local models through Ollama or llama.cpp, ensuring sensitive data never leaves a user’s network. Alternatively, it can be deployed on cloud infrastructure such as Alibaba Cloud Elastic Compute Service (ECS) or other VPS environments.
The assistant is designed for multi-platform integration, supporting channels including Discord, Slack, DingTalk, Feishu, and iMessage, enabling a single AI assistant to operate across multiple communication platforms.
Built on the AgentScope framework, CoPaw allows developers to create custom Python-based skills, enabling tasks such as database queries, web scraping, or enterprise workflow automation. It also includes cron-based scheduling for autonomous task execution and supports switching between models—including GPT-based, Llama, and Qwen models—without losing stored memory.
With flexible installation options, including one-line installers, Docker images, and pip installation, CoPaw aims to provide developers with a controllable and extensible AI assistant infrastructure.
