← Chat

HuggingChatHuggingChat

Hugging Face

开源聚合Open-source aggregation 前端开源Open-source frontend Apache-2.0Apache-2.0 免费Free 未备案Not filed partial lifecycle: active 核验Checked 2026-10-30

HuggingChat 是「只用开源模型」的聚合对话产品（HF 出品，当日 137 个模型+Omni 自动路由）：前端 chat-ui 开源（Apache-2.0）可自部署——它同时是产品，也是开源 chat 前端的参考实现。

HuggingChat is the "open-source models only" aggregator chat product (by Hugging Face, 137 models + Omni auto-routing on the day): the chat-ui frontend is open source (Apache-2.0) and self-deployable — it is both a product and the reference implementation of an open-source chat frontend.

适合与不适合When to use it, when not to

适合：只用开源模型、横排对比开源档、参考/自部署开源前端。不适合：要用闭源旗舰模型、要大陆直连。

Fits: open-source models only, side-by-side open-source comparison, referencing/self-deploying an open-source chat frontend. Not a fit: closed-source flagship models, direct mainland access.

中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-30

出品属地Origin: 国外出品foreign-made　国外主体（Hugging Face，美国；网信办公示名单未见对应备案主体）。foreign entity (Hugging Face, USA; no matching filing entity in the CAC registry).
国内使用Direct access: 直连不可达direct blocked　huggingface.co 大陆直连 000（网络不可达，2026-10-30；既有口径：HF 主域在大陆长期不可达，镜像站是另一话题）。huggingface.co measured 000 direct from mainland (network unreachable, 2026-10-30; long-standing HF unavailability; mirror sites are a separate topic).
备案状态CAC filing: 未见于备案名单not in filing list：未见于网信办公示名单（2026-10-30 检索百炼公示页与批次 PDF 镜像，未见 Hugging Face 条目）；该产品未面向大陆公众提供服务，备案不适用。not in the CAC public registry (Bailian disclosure page and batch PDF mirrors searched 2026-10-30; no Hugging Face entry); the product does not serve the mainland public, so filing does not apply.

备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.

固定坐标系Fixed coordinate system

8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.

模型与开放条件Model & openness: 开源模型聚合是它的本体形态：当日页内可见「Models 137」清单与 Omni 自动路由（官方文案「Omni automatically picks the best AI model… You can also choose from any available open source models」）—— 用户直接选开源模型，全部托管在 HF 推理侧。
Open-source model aggregation is its native form: the in-page list "Models 137" and Omni auto-routing were visible on the day (official copy: "Omni automatically picks the best AI model… You can also choose from any available open source models") — users pick open-source models directly, all hosted on HF's inference side.
运行位置Where it runs: 网页（huggingface.co/chat）；前端开源（chat-ui 仓 Apache-2.0，当日 ★11,004/pushed 2026-10-30）—— 可自部署（runner 类的 text-generation-webui 等是另一条本机路线）。托管版生成在 HF 侧。
Web (huggingface.co/chat); the frontend is open source (the chat-ui repo, Apache-2.0, ★11,004 / pushed 2026-10-30 verified via GitHub API same day) — self-deployable (the runner-class tools like text-generation-webui are a separate local route). The hosted version generates on HF's side.
本地文件Local files: 托管版对话历史存于账户/浏览器侧；自部署版数据完全在自己机器上（这是它与托管对话产品的结构性差别）。
On the hosted version, chat history lives on the account/browser side; on a self-deployed version data stays entirely on your own machine (the structural difference from hosted chat products).
关机后的任务Tasks after shutdown: 对话即时返回；异步任务形态未核验到官方声明。当日页内新增「ML Intern」入口（papers/training/spaces 向）。
Chat returns instantly; no official statement verified for an async-task form. A new "ML Intern" entry (papers/training/spaces-oriented) was on the page on the day.
工具与扩展Tools & extensions: 开源前端的 MCP 工具接入在官方 README 在册（经 OpenAI function calling 调用 MCP server 暴露的工具、可经 env 预配置受信 server、用户也能在界面自行添加并做健康检查，预配置示例含 Web Search (Exa)）；托管版按同一代码库运行，具体开关形态未逐条核验。
MCP tool integration in the open-source frontend is on record in the official README (tools exposed by MCP servers called via OpenAI function calling; trusted servers can be preconfigured via env, users can add their own in the UI with a health check; the preconfigured example includes Web Search (Exa)); the hosted version runs on the same codebase, its exact toggle form not verified item by item.
上下文与记忆Context & memory: 会话按对话组织、托管版历史存于账户/浏览器侧；跨会话记忆机制未核验到官方声明—— 开源前端的数据层完全归自部署者管辖，这是它与托管对话产品的结构性差别。
Chats organized per conversation; hosted history lives on the account/browser side. No official statement verified for cross-session memory — the open-source frontend's data layer is entirely under the self-deployer's control, the structural difference from hosted chat products.
权限与限制Permissions & limits: HF 账号可选（不登录也能用，功能受限）；数据口径按 HF 隐私页与所选模型许可双重管辖，未逐条核验。
HF account optional (usable without login, with reduced features); data terms are governed jointly by HF's privacy page and each chosen model's license, not verified item by item.
适合什么任务Best for: 适合：只用开源模型、要一个账号横排 137 个开源档、要参考/自部署开源 chat 前端。不适合：要用闭源旗舰模型（GPT/Claude 不在列）、要大陆直连（未备案）。
Fits: open-source models only, lining up 137 open entries under one account, referencing/self-deploying an open-source chat frontend. Not a fit: closed-source flagship models (GPT/Claude are not on the list), direct mainland access (not filed).

头号误解Top misconception

HuggingChat 只有开源模型——它与 Poe 同为聚合但模型池完全不同（Poe 含闭源 API）HuggingChat lists open-source models only — like Poe it aggregates, but the model pools are completely different (Poe includes closed APIs)
「前端开源可自部署」与「托管版跑 HF 推理」是两件事——自部署要自己接推理后端（挂 runner 类工具）"Frontend open source, self-deployable" and "the hosted version runs on HF inference" are two things — self-deploying means wiring your own inference backend (hang runner-class tools)
各模型许可不同（有的商用限制）——聚合界面不改变每个权重自身的许可条款Model licenses differ (some carry commercial restrictions) — an aggregation UI does not change each weight's own license terms

未知项清单Known unknowns

Omni 路由逻辑细节（未核验）
托管版数据留存与训练口径（未逐条核验）
跨会话记忆机制（未核验到官方声明）
托管版 MCP server 开关的具体形态（README 在册，托管界面未逐条核验）

Omni routing logic details (unverified)
Hosted-version data retention and training terms (not verified item by item)
Cross-session memory mechanism (no official statement verified)
Exact form of the hosted version's MCP server toggle (on record in the README; hosted UI not verified item by item)

证据来源Evidence sources

判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-30，内容更新日content updated on 2026-10-30。

类型Type	名称Name	链接Link
official	HuggingChat · 产品入口（2026-10-30 代理抓 308→200；Models 137/Omni/ML Intern 当日在册）HuggingChat · product entry (proxy-fetched 308→200 on 2026-10-30; Models 137 / Omni / ML Intern on record same day)	https://huggingface.co/chat
official	HuggingChat · Terms（产品条款，2026-10-30 代理抓 200）HuggingChat · Terms (product terms, proxy-fetched 200 on 2026-10-30)	https://huggingface.co/chat/terms
official	HuggingChat · Privacy（产品隐私页，2026-10-30 代理抓 200）HuggingChat · Privacy (product privacy page, proxy-fetched 200 on 2026-10-30)	https://huggingface.co/chat/privacy
repo	chat-ui · 官方仓（前端代码，Apache-2.0，2026-10-30 GitHub API 核验 ★11,004/pushed 2026-10-30）chat-ui · official repo (frontend code, Apache-2.0, GitHub API verified 2026-10-30: ★11,004 / pushed 2026-10-30)	https://github.com/huggingface/chat-ui
changelog	chat-ui · Releases（变更记录，2026-10-30 代理抓 200）chat-ui · Releases (change log, proxy-fetched 200 on 2026-10-30)	https://github.com/huggingface/chat-ui/releases
docs	chat-ui · 文档（2026-10-30 代理抓 200）chat-ui · docs (proxy-fetched 200 on 2026-10-30)	https://huggingface.co/docs/chat-ui/main/index

实测记录Measured runs

本档未跑对话任务集；能力结论来自官方文档，不评价回答质量。

实测建议（三条，都指向官方可读页没给的东西）：

Omni 自动路由的选择逻辑与可解释性（官方只说 picks the best）
托管版对话数据的留存与训练口径（HF 隐私页逐项看）
自部署 chat-ui 的推理后端接入成本（官方文档有路径，实操未测）

This site has not completed measured runs. The coding-oriented test protocol is at protocols/coding/_protocol.md (the coding task suite does not apply to chat products).

Suggested measurements (three, all pointing at things the official readable pages do not give):

Omni auto-routing's selection logic and explainability (the official copy only says "picks the best")
Hosted-version chat-data retention and training terms (read the HF privacy page item by item)
Cost of wiring inference backends to a self-deployed chat-ui (the official docs give the path; hands-on untested)