正在开发 Cellium-Agent构建模块化 Agent Runtime 与 AI 运行时自扩展工具。
目前专注于 Agent Runtime 的设计与实现。
关注方向:
- Agent Runtime
- Runtime Self-Extension
- Vision Systems
- Memory Architecture
- Context Engineering
- Long Running Tasks
- Desktop Automation
相比模型能力本身,更关注如何构建能够长期运行、持续演化、自主执行的 Agent 系统。
模块化桌面 Agent 框架。
核心设计原则:
- High Cohesion
- Loose Coupling
- Provider Agnostic
- Runtime Self-Extension
- Observable Execution
目标是构建面向长期运行任务的 Agent Runtime。
聚焦于:
- Agent Runtime
- Tool System
- Memory System
- Vision System
- Context Management
- Desktop Automation
- 屏幕感知能力
- 视觉上下文管理
- 多模态输入融合
- 视觉事件抽象
- 长期记忆增强
- 检索优化
- 上下文压缩
- 记忆隔离机制
- 运行时自扩展
- 长时任务执行
- Agent 生命周期管理
Agent ≠ Model
Model 提供智能
Runtime 提供执行能力
Memory 提供连续性
Vision 提供环境感知
Tools 连接现实世界
运行时自扩展 Agent 框架
GitHub: https://github.com/Cellium-Project/Cellium-Agent
