← Safety

HeliconeHelicone

Helicone

AI 网关AI gateway 观测observability 日志logs 用量usage 多家模型multi-provider partial lifecycle: active 核验Checked 2026-11-06

核验标记表示资料完整度，不是产品评分。下列未确认项可能影响选型；请按对应来源逐项确认。The confidence mark describes evidence completeness, not a product score. Unconfirmed items below may affect your choice; check them against the relevant sources.

未确认的信息Information not confirmed

许可证类型（当日未取到官方表述）
是否提供自托管及其功能范围
请求/响应日志的留存位置、时长与删除方式
数据是否用于任何训练或改进用途
团队权限分级、审计日志与导出能力
计费口径与限额（价格数字不进档案）

Licence type (no official statement captured on the check date)
Whether self-hosting exists and what it includes
Where request/response logs are stored, for how long, and how they are deleted
Whether logged data is used for any training or product improvement
Team permission tiers, audit logs and export capabilities
Billing and quota terms (prices are not recorded in entries)

English summaries and interface labels are available. Detailed source notes below are currently in Chinese.

Helicone 是放在你的应用与多家模型之间的网关：官方描述它覆盖 100+ 模型，内置自动日志、观测与失败回退，密钥由平台侧托管、用量在控制台里看。它解决的是「调用集中管理」，不是「模型质量判断」。

适合与不适合When to use it, when not to

适合：需要一家以上模型接入、想把日志/用量/失败回退收在一处，且接受请求经第三方平台。不适合：数据不允许离开自己环境的场景；或要求自托管但你还没核到官方部署说明的场景。

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 2026-11-06

出品属地Origin: 国外出品foreign-made　文档与更新记录均为境外站点，当日直抓可达。Docs and changelog are hosted outside the mainland; both were reachable on the check date.
国内使用Direct access: 大陆可直连direct OK　docs.helicone.ai 与 www.helicone.ai/changelog 直抓 200（2026-11-06）。docs.helicone.ai and www.helicone.ai/changelog returned 200 direct on 2026-11-06.
备案状态CAC filing: 备案未核验filing unknown：备案与境内服务口径当日未核到；页面可达不等于境内可用。Filing and in-country availability not verified on the check date; reachability is not availability.

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: 官方文档把自己描述为「AI Gateway」，称覆盖 100+ 模型；模型密钥由平台侧托管，用户在账户里充值后调用。
The official docs describe it as an "AI Gateway" covering 100+ models, with provider keys held on the platform side after you add credits to your account.
运行位置Where it runs: 托管网关形态：请求经由平台，官方称调用记录在 Requests 页面「数秒内」出现。是否提供自托管，当日抓取的那页未给说明。
A hosted gateway: requests pass through the platform, and the docs say entries appear in the Requests tab "within seconds". Whether self-hosting exists is not stated on the page captured on the check date.
本地文件Local files: 用托管网关意味着请求与响应经过平台；本地文件是否留存、留存多久，当日未核到具体政策。
Using a hosted gateway means requests and responses traverse a third party. Whether local files are retained, and for how long, was not verified for this entry.
关机后的任务Tasks after shutdown: 网关常驻服务，官方列出 automatic logging、observability 与 fallbacks；后台任务的持久队列能力面未逐条核验。
A standing service; the official page lists automatic logging, observability and fallbacks. Persistent background queue behaviour was not verified item by item.
工具与扩展Tools & extensions: 官方描述的是网关与观测能力；作为工具扩展生态的一部分（SDK、代理配置）当日未逐条核验。
The stated surface is gateway routing and observation; SDK and proxy configuration details as an extension ecosystem were not verified item by item.
上下文与记忆Context & memory: 上下文由你接入的模型决定；平台对请求体的记录范围、截断与保留窗口当日未核到。
Context length belongs to whichever model you call; the platform's logging scope, truncation and retention windows were not verified.
权限与限制Permissions & limits: 账户体系内管理密钥与用量（官方写「maintains the keys for you」）；团队分级、审计日志与数据导出权限当日未逐条核验。
Keys and usage sit inside the account model (the docs say it "maintains the keys for you"). Team tiers, audit logs and export permissions were not verified item by item.
适合什么任务Best for: 适合：想用一个入口统一多家模型调用、集中看用量与失败，并接受请求经第三方平台。不适合：数据不允许离开自己环境的场景；需要自托管且必须核到官方部署说明的场景。
Fits: one entry point for several providers, consolidated logs, usage and failures, with requests leaving your environment accepted. Does not fit: data that must not leave your environment, or a self-hosting requirement you have not yet matched against an official deployment statement.

头号误解Top misconception

把「统一入口」当成「统一能力」——各家模型的参数与行为差异不会因网关消失
只看观测面板方便，没确认请求体里带了什么敏感内容就送出去
把平台侧托管密钥当成自己不需要做密钥轮换与最小权限

证据来源Evidence sources

最近核验记录Latest recorded check 2026-11-06，内容更新日content updated on 2026-11-06。单项复核范围见正文或来源说明，不代表所有字段在同一天重新核验。See the text or source notes for item-specific scope; this does not mean every field was rechecked on that date.

类型Type	名称Name	链接Link
docs	Helicone · 官方文档（AI Gateway · automatic logging, observability and fallbacks；100+ models）Helicone · official docs (AI Gateway · automatic logging, observability and fallbacks; 100+ models; 200 direct on 2026-11-06)	https://docs.helicone.ai
changelog	Helicone · 更新记录Helicone · changelog (200 direct on 2026-11-06)	https://www.helicone.ai/changelog

官方口径与未核到项Official record and unverified items

本档未做延迟、可用性或回退生效条件的测试；「100+ 模型」「数秒内出现」等表述是官方页原文，本站不重复其数字作为评价依据，也不据此评价同类产品。

This entry has not completed measured runs on latency, availability or failover conditions. "100+ models" and "within seconds" are quoted from the official page; the site does not treat those figures as evidence for a ranking, and draws no comparison against similar products.

第一次怎么用

先过一遍你要送出去的数据：客户姓名、订单号、密钥、源码片段。任一项不能离开自己环境，就先别用托管网关。
接入时把「谁在调用哪个模型、失败会怎样」写清楚：网关只保证集中，不保证行为一致；回退条件与超时阈值要显式设定并演练一次。
密钥仍在你的账上：平台代管不等于免轮换，按你的离职与泄露预案设定轮换周期。
上线后核对日志范围：确认记录里没混进不该长期留存的内容，再决定保留多久、怎么导出。