Show HN: Nully – FOSS AI chat without the bloat Show HN：Nully——无冗余的开源AI聊天工具

TL;DR

• Nully is a lightweight, self-hostable AI chat application built with Go that integrates directly with the OpenRouter API.
• The tool emphasizes privacy by design, ensuring user data and API keys remain local to the device with no central server tracking.
• It offers significant performance advantages over major competitors, boasting faster load times, lower resource consumption, and reduced memory usage.
• Features include support for hundreds of models, file attachments, web search capabilities, and easy deployment via Docker or direct execution.

Why It Matters

This release highlights a growing trend toward decentralized, privacy-first interfaces for accessing large language models, addressing user concerns over data sovereignty and surveillance capitalism in AI interactions. For developers and power users, Nully provides a highly efficient alternative to bloated commercial platforms, demonstrating that robust AI chat experiences can be delivered with minimal infrastructure overhead. It serves as a practical example of how leveraging existing API aggregators like OpenRouter can simplify the development of secure, multi-model client applications.

Technical Details

• Architecture: Built using Go, resulting in a single binary executable with no external database dependencies or complex build steps required for basic operation.
• Privacy Mechanism: Implements a direct-to-API communication pattern where messages bypass Nully’s servers entirely, going straight from the browser to OpenRouter and the selected model provider.
• Performance Metrics: Benchmarks indicate a Largest Contentful Paint (LCP) of 0.2s compared to ChatGPT’s 0.6s, with resource loading at 210kB versus 11.8MB+ for competitors, and peak memory usage of 0.9MB against 32MB+.
• Deployment Options: Supports self-hosting via go run or Docker containers, alongside a hosted web version at nully.chat, allowing flexible integration into various workflows.
• Feature Set: Includes streaming responses, support for images/PDFs, web search integration, portable history export, and theme customization.

Industry Insight

The success of lightweight, privacy-centric tools like Nully suggests that users are increasingly demanding control over their AI interactions, pushing vendors to prioritize data minimization and transparency. Developers should consider adopting modular, API-agnostic designs that allow users to swap underlying model providers without changing the interface, thereby increasing flexibility and reducing vendor lock-in. Additionally, optimizing for extreme efficiency in client-side applications can significantly enhance accessibility and performance, particularly in low-bandwidth or resource-constrained environments.

TL;DR

• Nully 是一款轻量级、开源的 AI 聊天应用，专为 OpenRouter 设计，强调极简主义与高性能。
• 采用“自带 API Key”模式，数据完全存储在用户本地设备，不经过 Nully 服务器，确保隐私安全。
• 性能表现卓越，资源加载仅 210kB，内存占用约 0.9MB，显著优于 ChatGPT、Claude 等主流竞品。
• 支持自托管部署（Go/Docker），具备附件上传、网页搜索及多主题适配功能，无需注册账号。

为什么值得看

对于注重数据隐私和系统资源效率的开发者及高级用户而言，Nully 提供了一个无需云端依赖即可访问数百种大模型的极佳方案。其极致的轻量化设计展示了在保持丰富功能的同时，如何大幅降低 Web 应用的资源开销，为构建高效、私有的 AI 前端界面提供了优秀的参考范例。

技术解析

• 架构与隐私机制：应用采用纯前端或极简后端架构，消息直接从浏览器发送至 OpenRouter 及选定的模型提供商，Nully 不存储任何中间数据。API Key 仅保存在本地，并通过配置限制其只能发送给 OpenRouter，防止密钥泄露。
• 性能基准对比：根据 2026 年 7 月的测试数据，Nully 的最大内容绘制时间（LCP）仅为 0.2s，资源加载大小 210kB，峰值内存使用 0.9MB。相比之下，ChatGPT 的资源加载为 15.1MB，内存占用 32MB；Claude 内存占用高达 94MB。
• 部署灵活性：项目基于 Go 语言开发，无数据库依赖，无构建步骤。支持通过 go run 直接运行，或通过 Docker 构建微小容器进行自托管，也可直接使用托管版 nully.chat。
• 功能特性：支持图片、PDF 等多媒体附件发送，集成模型网页搜索能力，允许用户调整模型的“思考深度”，并支持聊天记录的本地导出与导入，实现历史便携性。

行业启示

• 隐私优先的 AI 交互范式：随着用户对数据隐私关注度提升，去中心化、本地化存储的 AI 客户端需求增长。Nully 证明了无需建立庞大后端基础设施也能提供安全、合规的 AI 服务体验。
• 极致轻量化的工程价值：在 AI 应用日益臃肿的背景下，Nully 展示了通过精简技术栈（如 Go + 静态资源）实现毫秒级加载和低内存占口的可行性，有助于在低带宽或低端设备上提供更好的用户体验。
• 开放生态的聚合器机会：通过整合 OpenRouter 等模型路由平台，单一前端应用即可接入海量异构模型，降低了用户切换不同 AI 服务的门槛，这种“前端聚合+后端路由”的模式具有广泛的商业和技术推广潜力。