隐私与广告选择
Git-Stars 会使用必要存储来保障网站运行。可选分析和广告测量脚本默认不加载，只有在你同意后，Google 等合作伙伴才可能按要求使用 Cookie 或类似标识符。 隐私政策
这个仓库详情页已经具备足够的原创分析、来源证据和同类对比，可以作为 Git-Stars 的可索引评测页。
100
评测分
解决的问题
Needle solves the problem of running capable AI agents on resource-constrained devices without cloud dependency, offering a tiny model that outperforms larger ones (e.g., FunctionGemma-270m, Qwen-0.6B) on single-shot function calling. It enables local finetuning on consumer hardware, reducing latency and privacy concerns compared to cloud-based solutions.
部署现实
当前可用的安装信号是：git clone https://github.com/cactus-compute/needle.git && cd needle && source ./setup。这只能作为起点，正式使用前建议让 Codex 或 Claude 检查 README、环境变量、运行时版本和部署目标。
商业可用性
当前记录的许可证是 MIT。这是重要的商业使用信号，但仍应核验许可证全文、依赖许可证、模型/API 条款和商标限制。
能力上限
Developers can build personal AI assistants that execute function calls (e.g., weather queries, tool invocations) directly on-device with low latency. Real-world use cases include smart glasses that fetch information hands-free, smartwatches that control IoT devices, and mobile apps that process user commands offline. The ceiling includes complex multi-step agentic workflows, though the model is optimized for single-shot tasks and may struggle with conversational depth.
Git-Stars 是独立产品，不隶属于 GitHub 或该项目。 分析可能由 AI 辅助生成，依据公开仓库元数据和 README 的短摘要。 我们不镜像完整 README、文档、Issue 或社媒评论。
cactus-compute/needle 被追踪为 Python 项目，主要属于 AI Agent, LLM Tool 方向。这个评估结合公开 GitHub 元数据、分类信号、短来源摘要和 Git-Stars 编辑规则，而不是复制项目文档。
增长检查：该仓库目前有 6.1k Star，今日 +547，本周 +2.5k，本月 +0。这些窗口用于区分持续采用信号和短期曝光峰值。
维护检查：当前活跃度为 一般；最近一次推送距今 60 天，未关闭 Issue 为 25，约占总 Star 的 0.41%。这只是采用信号，不替代工程尽调。
采用检查：405 Fork 和 6.1k Watcher 反映项目被复用和关注的程度。许可证信号：MIT。商业或内部使用前请核验许可证兼容性。
适用判断：当你需要「AI 原型、LLM 工作流和 Agent 类应用」时，这个项目更值得评估；如果「需要法律审查、安全审计或生产 SLA 保证」，则需要谨慎。
来源检查：Git-Stars 当前为这份报告保留了 2 个明确来源引用，近期增长信号为 2.5k。最终安装、安全和版本信息仍应以原始 GitHub 仓库为准。
原创分析
充分已经具备问题、能力和结论层面的原创判断。
来源
充分已附带仓库元数据和 README/来源引用。
许可证
充分已记录 MIT，可作为评估依据。
维护
充分仓库元数据中可以看到近期活跃信号。
同类对比
充分已有足够相近项目用于对比。
热度
6.1k 星标
复用
405 复刻
关注
6.1k 关注者
维护
moderate
许可证
MIT
未解决 Issue
25
Needle is a 26m parameter Simple Attention Network distilled from Gemini 3.1, designed for efficient single-shot function calling on consumer devices. It runs at high speed on Cactus infrastructure and can be finetuned locally on a Mac or PC.
核心特性
- Distilled from Gemini 3.1 into a 26m parameter model with 6000 toks/sec prefill and 1200 decode speed. - Fully open weights and dataset generation, with easy finetuning via web UI or CLI. - Outperforms larger models (FunctionGemma-270m, Qwen-0.6B) on single-shot function call tasks.
AI Agent
Agent frameworks, autonomous workflows, and tool-use systems
LLM Tool
Libraries and tools for LLM apps, RAG, prompts, and evals
git clone https://github.com/cactus-compute/needle.git && cd needle && source ./setup63
健康评分
一般
提交活跃度
Feb 24, 2026
创建于
Aug 15, 2026
最近提交
每日同步积累数据后，Star 历史将在此显示。
复刻
未解决
关注者
cactus-compute
GitHub 主页obra/superpowers
An agentic skills framework & software development methodology that works.
affaan-m/ECC
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
NousResearch/hermes-agent
The agent that grows with you
mattpocock/skills
Skills for Real Engineers. Straight from my .agents directory.
multica-ai/andrej-karpathy-skills
A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.
✓
许可证
✓
Fork 情况
~ 一般
维护情况
解决的问题
Needle solves the problem of running capable AI agents on resource-constrained devices without cloud dependency, offering a tiny model that outperforms larger ones (e.g., FunctionGemma-270m, Qwen-0.6B) on single-shot function calling. It enables local finetuning on consumer hardware, reducing latency and privacy concerns compared to cloud-based solutions.
能力范围
Developers can build personal AI assistants that execute function calls (e.g., weather queries, tool invocations) directly on-device with low latency. Real-world use cases include smart glasses that fetch information hands-free, smartwatches that control IoT devices, and mobile apps that process user commands offline. The ceiling includes complex multi-step agentic workflows, though the model is optimized for single-shot tasks and may struggle with conversational depth.
结论
Needle is ideal for developers building lightweight, on-device AI agents for single-shot function calling, especially in privacy-sensitive or low-latency applications. It should be avoided for tasks requiring deep conversational understanding or broad general knowledge. The key trade-off is extreme efficiency and specialization at the cost of versatility and scope.