终端编程 Agent(最接近 Claude Code)
1. OpenCode ⭐ 165k+
- 协议: MIT | 语言: Go/TypeScript
- 免费: 完全免费开源,支持 75+ LLM 提供商
- 特色:
- 事实上的开源 Claude Code 替代品
- 本地优先架构,支持本地模型(Ollama 等)
- Provider-agnostic:同一会话可切换 Claude/Gemini/GPT/本地模型
- 精美 TUI 界面,支持桌面端和 IDE 扩展
- 中国讨论度: 低,国内媒体很少报道
- 🔗
2. Pi ⭐ 54k+
- 协议: MIT | 语言: Python
- 作者: Armin Ronacher(Flask/Jinja2 作者)
- 免费: 完全免费开源
- 特色:
- 系统提示 < 1,000 tokens,极简设计
- "Lazy Skills" 机制:技能按需加载
- 专为 fork 和二次开发设计
- 增长速度惊人(短时间内突破 54k stars)
- 中国讨论度: 极低
- 🔗
3. Crush ⭐ 25k+
- 协议: FSL(2年后转MIT)| 语言: Go
- 团队: Charm(Bubble Tea 团队)
- 免费: 免费使用
- 特色:
- 终端美学标杆,极其流畅的 TUI
- 多模型支持,会话中可切换模型
- 原生 LSP 和 MCP 支持
- 适合追求终端体验的开发者
- 中国讨论度: 低
- 🔗
4. Qwen Code ⭐ 25k+
- 协议: Apache-2.0 | 语言: TypeScript
- 免费: 免费,配合 Qwen 模型有免费额度
- 特色:
- 阿里巴巴出品,Gemini CLI 的开源 fork
- Gemini CLI 6月停服后,这是其开源延续
- 专门优化 Qwen-Coder 模型
- 国内用户访问友好
- 中国讨论度: 中等(但远低于其价值)
- 🔗
---
通用 Agent 框架
5. OpenClaw
- 协议: 开源 | 免费: 完全免费,云版待定
- 特色:
- 自托管 AI Agent,50+ 原生集成
- 零外部 API 调用,隐私优先
- 连接 Slack/GitHub/Notion 等无需第三方 API
- 支持 Docker 部署
- 中国讨论度: 低
- 🔗
6. Agno
- 协议: 开源 | 语言: Python
- 免费: 开源免费,平台版 $99/月起
- 特色:
- 2 微秒 Agent 运行时间,极致轻量
- 内置记忆、存储、多模态工具
- 生产级 Python 框架
- 适合需要高性能的场景
- 中国讨论度: 极低
- 🔗
7. Hermes Agent ⭐ 60k+
- 协议: 开源
- 免费: 开源免费
- 特色:
- 2个月内突破 60k stars,增速最快
- 持久多层记忆:长期语义记忆 + 工作记忆 + 情景日志
- 云优先架构,独立于本地设备
- 30天学习期后能理解用户工作模式
- 与小米 MiMo V2 Pro 合作
- 中国讨论度: 中等(小米合作有报道,但产品本身讨论少)
- 🔗
---
MCP 生态工具
8. MCPX (IBM)
- 协议: 开源
- 免费: 开源免费
- 特色:
- MCP 网关,统一管理多个 MCP 服务器
- Tool Groups:不同团队看到不同工具子集
- Agent 访问控制 + 实时 Prometheus 指标
- 支持 Cursor/Claude Code/VS Code/Copilot 等
- 中国讨论度: 极低
- 🔗
9. ContextForge (IBM)
- 协议: 开源
- 免费: 开源免费
- 特色:
- 联邦化 AI 网关,跨多集群 Kubernetes
- 支持 MCP/A2A/REST-to-MCP/gRPC-to-MCP
- 40+ 插件
- OpenTelemetry 追踪
- 中国讨论度: 极低
- 🔗
---
多 Agent 编排
10. Microsoft Conductor ⭐ 新项目
- 协议: MIT | 版本: v0.1.1
- 免费: 完全免费开源
- 特色:
- 微软出品,GitHub Copilot SDK + Anthropic Agents SDK
- YAML 定义工作流 + Web 仪表板
- 适合企业级多 Agent 场景
- 中国讨论度: 极低
- 🔗
11. Agent Skills (by Addy Osmani) ⭐ 43.8k+
- 协议: 开源
- 免费: 完全免费
- 特色:
- 23 个生产级工程技能
- 7 个斜杠命令覆盖完整开发生命周期
- 编码 Google 工程文化(Hyrum's Law、trunk-based development)
- 渐进式披露设计
- 中国讨论度: 低
- 🔗
显示更多
Behind the MiMo API Price Reduction:
The deepest price cut, up to 99%, is for Input (Cache Hit). The core reason is our inference framework now supports hierarchical KV cache optimization for SWA. Production inference engine tests show this optimization increases cached token capacity by 5x, equivalent to an 80% reduction in caching costs. Combined with Cache Read Overlap among multiple Full Attention modules in the Hybrid model, actual costs are further reduced.
Prices for Input (Cache Miss) and Output are also reduced by 60%-80%. This mainly benefits from the extreme 1:7 Full:SWA sparsity ratio brought by the model architecture (the prefill compute of the 70-layer MiMo-V2.5-Pro roughly equals a 10-layer GQA model). This kept our original inference costs well below the industry average, naturally leaving a 2x-3x profit margin in pricing. This price adjustment simply reflects our decision to pass these structural cost efficiencies directly to developers.
Operating at these newly reduced API prices, our production inference engine is running at near full capacity, and we can still essentially break even. We previously advised LLM companies not to "blindly cut prices" precisely because very few model architectures and inference optimizations can keep API costs from running at a loss. If more architectures that save compute and KV cache emerge, along with better inference Infra to drive down API costs, this will form an excellent virtuous cycle in the industry.
More crucially, affordable, high-performance model APIs will drive real, sustained, and at-scale inference demand. This upstream demand pulls forward the development of the entire AI infrastructure chain—including chips, servers, optical transceivers, PCBs, liquid cooling, power, energy storage, and data centers—serving as a strategic fulcrum for a systemic revaluation of AI hardware. In the long run, this injects more affordable and accessible compute into both training and inference pipelines, accelerating the parallel evolution of global AGI across multiple regions and technical routes.
For more technical details, we will release a detailed Blog post later.
显示更多
I did a mini test on some tier 2 models I have. The test is to generate a macOS app that displays a Siri Remote device according to a product image I provided.
Xiaomi MiMo is at a disadvantage in this situation since it doesn't support the image function. Moreover, it was the slowest of the four. It took 10 min just to complete this interface. It is excusable considering it's free($0.01) for now.
Gemini's performance among these four models is actually quite impressive. It’s not as bad as others have complained.
显示更多