AI Intel

Cloudflare has introduced temporary accounts, enabling agents to deploy services before a full signup process.

Deno Desktop has been released, signifying the expansion of the Deno ecosystem into desktop application development.

A logging bug in Codex has been identified that could potentially write terabytes of data to local SSDs.

Benchmarks indicate that the Gemma 4 QAT 31B model responds more effectively to KV cache quantization, yielding better results.

An article provides a comparison and analysis of the GLM 5.2 and Opus models, discussing their performance and application differences.

A Reddit community discussion has started to debate and identify the best local AI agents currently available.

Skybridge, a full-stack open source React framework for MCP Apps, has been launched on Product Hunt.

Agent 37 Cloud 在 Product Hunt 上线，允许用户为每位客户部署定制的 Hermes 或 OpenClaw 代理。

Ollama 发布 v0.30.9 版本，重点更新包括支持 Cohere2Moe 架构，并修复了 LFM2 解析器和编码代理输出问题，提升了本地 LLM 运行的兼容性和稳定性。

Reddit 社区分享了一份详细的本地 LLM 推理优化指南，涵盖 VRAM 适配、KV 缓存、MoE 部署、MTP 和 CPU 调优等实践经验。