跳到主要内容Skip to main content
← Safety

LangWatchLangWatch

LangWatch
LLMOpsLLMOps 追踪tracing 评测evaluation 提示管理prompt management 可自托管self-hostable 开源核心open-source core 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

  • 托管版与自托管版的功能差异清单
  • 托管版的留存位置、删除时限与训练用途条款
  • 支持的模型与框架接入全清单
  • 自定义评判器、外部工具接入的确切边界
  • trace 的长度上限、采样策略与保留窗口
  • 托管版账号与团队权限分级
  • Feature delta between the hosted product and the self-hosted core
  • Retention location, deletion deadlines and training use on the hosted plan
  • Full list of supported models and frameworks
  • Exact boundaries of custom judges and external tool integrations
  • Trace size limits, sampling policy and retention window
  • Account and team permission tiers on the hosted product

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

LangWatch 是把线上调用、评测与提示版本收在一处的 LLMOps 平台:官方列出的能力是追踪每一次模型调用、工具使用与用户交互,并在其上跑评测与代理测试。开源核心可自托管(文档写为 Apache-2.0),另提供托管版。

LangWatch 是把线上调用、评测与提示版本收在一处的 LLMOps 平台:官方列出的能力是追踪每一次模型调用、工具使用与用户交互,并在其上跑评测与代理测试。开源核心可自托管(文档写为 Apache-2.0),另提供托管版。

适合与不适合When to use it, when not to

适合:需要把 trace、评测与提示变更放在一起复盘,同时希望保留自托管退路的团队。 不适合:只想要一个现成基准分数;或对「调用记录是否离开自己环境」有硬性限制却打算用托管版的场景。

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 官方文档与更新记录为境外站点,当日直抓可达(docs 由 docs.langwatch.ai 跳转至 langwatch.ai/docs/)。Docs and changelog are hosted outside the mainland; both were reachable on the check date (docs.langwatch.ai redirects to langwatch.ai/docs/).
国内使用Direct access
大陆可直连direct OK langwatch.ai/docs/ 与 langwatch.ai/changelog 直抓 200(2026-11-06)。langwatch.ai/docs/ and langwatch.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

官方文档把它定位为 LLMOps 平台,追踪「每一次模型调用、工具使用与用户交互」;被测模型由你接入,平台不自带模型。具体兼容清单当日未逐条核到。

The docs position it as an LLMOps platform that tracks "every LLM call, tool usage, and user interaction"; the models under test are yours to connect, the platform does not ship a model. The full compatibility list was not verified item by item on the check date.

运行位置Where it runs

两条运行位置:官方托管,或自托管开源核心(文档明写 Apache-2.0 core 可自托管)。托管版与自托管版的功能差异当日未逐条核到。

Two placements: the hosted product, or self-hosting the open-source core (the docs state the core is Apache-2.0 and self-hostable). Feature differences between hosted and self-hosted were not verified item by item.

本地文件Local files

自托管时数据在你自己的存储里;用托管版时追踪数据离开本机——留存位置与删除方式需按当日条款另行确认,本档案未核到具体政策。

Self-hosting keeps traces in your own storage; on the hosted product trace data leaves your machine. Retention location and deletion terms were not verified for this entry.

关机后的任务Tasks after shutdown

追踪与评测以在线服务形式持续运行;离线批处理与定时任务的能力面未逐条核验。

Tracing and evaluation run as a continuous service; the offline batch and scheduled-task surface was not verified item by item.

工具与扩展Tools & extensions

官方列出的能力包含 tracing、evaluations、agent testing、prompt management;能否接入自定义评判器与外部工具,当日正文未逐条说明。

The official list of capabilities is tracing, evaluations, agent testing and prompt management; whether custom judges and external tools can be plugged in is not spelled out on the pages checked.

上下文与记忆Context & memory

上下文由被测应用决定;平台侧对 trace 长度、采样与保留窗口的限制未逐条核验。

Context length belongs to the application under test; platform-side limits on trace size, sampling and retention windows were not verified.

权限与限制Permissions & limits

托管版有账号与团队权限体系(具体分级未核到);自托管版权限等同你所部署环境的访问控制。

The hosted product has an account and team permission model (tiering not verified); self-hosted, permissions equal whatever your deployment grants.

适合什么任务Best for

适合:要把线上调用、评测与提示版本放在一处看,且希望保留自托管退路。 不适合:只想要一次性基准分数;或不能接受追踪数据离开自己环境的场景(那需要走自托管并自行核实功能边界)。

Fits: keeping live calls, evaluations and prompt versions in one place while retaining a self-hosting path. Does not fit: anyone wanting only a ready-made benchmark score, or anyone who cannot let trace data leave their environment and plans to use the hosted product.

头号误解Top misconception

  • 把「能追踪」当成「能判定质量」——评测判法与样本仍要你自己定义
  • 只看开源核心可自托管,就假设托管版与自托管版功能一致
  • 允许平台记录完整输入输出前,没先确认里面是否含客户数据与密钥

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
docsLangWatch · 官方文档(tracing / evaluations / agent testing / prompt management,自托管说明)LangWatch · official docs (tracing / evaluations / agent testing / prompt management, self-hosting note; 200 direct on 2026-11-06)https://langwatch.ai/docs/
changelogLangWatch · 更新记录LangWatch · changelog (200 direct on 2026-11-06)https://langwatch.ai/changelog

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

本档未做拦截率、评测一致性或性能测试;能力描述取自当日可达的官方文档与更新记录原文,本站不重复其数字,也不据此评价同类产品。

This entry has not completed measured runs on interception rates, evaluation consistency or performance. Capability wording comes from the official docs and changelog reachable on the check date; the site does not restate their figures as an evaluation, and draws no comparison against similar products.

第一次怎么用

  1. 先决定运行位置:能接受托管版就核对留存与删除条款;不能接受就按文档自托管开源核心,并逐项确认你需要的是否在核心内。
  2. 接入时只送必要字段:把客户姓名、订单号、密钥这类内容在进入 trace 前处理掉,事后清理比事前控制难。
  3. 评测从你自己的样本集开始:先写判法(哪些能机器判、哪些必须人判),再挂到线上 trace 上,别用平台默认指标替代验收标准。
  4. 每次改动记录三件东西:提示版本、模型标识、评测样本集版本;缺一件,下一轮的对比就失去解释力。

相关条目Related entries

  • HeliconeHelicone同属「评测与观测」线:本档案只写自己在册的官方口径,对照请看那份。Same line: Helicone (in evaluation and observability); this entry states only what its own official sources support.
  • OpenAI EvalsOpenAI Evals同属「评测与观测」线:本档案只写自己在册的官方口径,对照请看那份。Same line: OpenAI Evals (in evaluation and observability); this entry states only what its own official sources support.
  • InspectInspect同属「评测与观测」线:本档案只写自己在册的官方口径,对照请看那份。Same line: Inspect (in evaluation and observability); this entry states only what its own official sources support.

本页由This page is generated from the ai-handbook 数据层生成(CC BY 4.0)。data layer (CC BY 4.0). 方法论与坐标系定义见仓库内For the methodology and coordinate definitions, see METHODOLOGY.md.