An AI feature passes review, ships, and works fine in the demo. A week later the support queue fills up. The assistant is quoting a refund policy that doesn't exist, an agent is looping on the same tool call until it times out, and your model bill has tripled. You open your logs to find a request ID, a 200 status code, and nothing about what the model actually did.一项 AI 功能通过了评审并上线，在演示中运行良好。一周后，支持队列爆满。助手引用了一个不存在的退款政策，代理程序在同一个工具调用中陷入循环直到超时，而你的模型账单已经翻了三倍。你打开日志，只看到一个请求 ID、一个 200 状态码，却没有任何关于模型实际执行了什么的信息。
That gap is the reason AI observability exists. Traditional application monitoring was built to answer "is the service up and fast?" It tracks latency, error rates, and throughput, and it does that well. But none of those metrics tell you whether an answer was correct, why an agent chose the wrong tool, which retrieved document poisoned a response, or how a single user session burned through 40,000 tokens. AI systems fail in ways a green dashboard never shows.这种差距正是 AI 可观测性存在的原因。传统的应用程序监控旨在回答“服务是否正常且快速？”它跟踪延迟、错误率和吞吐量，并且做得很好。但这些指标都无法告诉你答案是否正确、代理为何选择了错误的工具、哪个检索到的文档污染了响应，或者单个用户会话是如何消耗掉 40,000 个 Token 的。AI 系统的故障方式是绿色仪表盘永远无法显示的。
This post walks through what an AI observability platform solves, the benefits worth paying for, the capabilities to check for, and the top platforms to evaluate in 2026.本文将探讨 AI 可观测性平台解决了什么问题、值得付费的优势、需要检查的功能，以及 2026 年值得评估的顶级平台。
The problem AI observability platforms solve#AI 可观测性平台解决的问题
LLM and agent applications are non-deterministic, multi-step, and expensive to run. A single user request can fan out into a chain of model calls, tool invocations, retrieval queries, validation steps, and downstream API calls. When the output is wrong, the failure could sit anywhere in that chain, and most of those steps leave no trace in a conventional logging setup.LLM 和代理应用程序具有非确定性、多步骤且运行成本高昂的特点。单个用户请求可能会扩展为一系列模型调用、工具调用、检索查询、验证步骤和下游 API 调用。当输出错误时，故障可能存在于该链条的任何位置，而大多数此类步骤在传统的日志记录设置中都不会留下痕迹。
Four problems show up often in production:生产环境中经常出现四个问题：
-
Silent quality failures. The system returns a confident answer that is wrong. Hallucinations, off-topic responses, and broken structured outputs all pass a status-code health check.静默质量故障。系统返回了一个自信但错误的答案。幻觉、离题的响应以及损坏的结构化输出都能通过状态码健康检查。
-
Invisible agent behavior. Multi-step agents make decisions you never see. Without a trace of the reasoning loop, a tool-selection bug looks identical to a model that's "just being weird."不可见的代理行为。多步骤代理会做出你从未察觉的决策。如果没有推理循环的追踪，工具选择错误看起来就和模型“只是表现得有点奇怪”一模一样。
-
Runaway cost. Token usage and model spend scale with traffic and prompt size in ways that are hard to predict. Teams routinely discover a 10x cost spike after the invoice arrives.失控的成本。Token 使用量和模型支出随着流量和提示词大小的增加而增加，且难以预测。团队经常在收到账单后才发现成本激增了 10 倍。
-
Root causes outside the model. A slow database query, a rate-limited API, or a malformed retrieval result often surfaces as what looks like a model problem. Tools that only watch the LLM call miss the actual cause.模型之外的根本原因。缓慢的数据库查询、受限的 API 或格式错误的检索结果通常表现为模型问题。只监控 LLM 调用的工具会错过真正的诱因。
An AI observability platform captures the full execution of every request so you can see what happened, in what order, and why, then attach quality scores and cost data to that record.AI 可观测性平台会捕获每个请求的完整执行过程，以便你查看发生了什么、顺序如何以及原因，然后将质量评分和成本数据附加到该记录中。
The benefits of an AI observability platform#AI 可观测性平台的优势
Teams adopt AI observability for concrete returns, not for nicer dashboards.团队采用 AI 可观测性是为了获得实际回报，而不是为了更美观的仪表盘。
Faster debugging. When you can replay an entire request as a single trace, time to resolution drops from hours of guesswork to minutes of knowing. You stop reproducing bugs by hand and start understanding what already happened in production.更快的调试。当你能将整个请求重放为单个追踪（Trace）时，解决问题的时间将从数小时的猜测缩短为几分钟的明确判断。你不再需要手动重现错误，而是开始理解生产环境中已经发生的事情。
Quality you can measure. Pairing traces with evaluations turns "the output felt off" into a number you can track over time. You catch quality regressions the same way you catch latency regressions, before users do.可衡量的质量。将追踪与评估配对，可以将“输出感觉不对”转化为你可以随时间跟踪的数字。你可以在用户发现之前，像捕获延迟回归一样捕获质量回归。
Cost control. Per-request token and spend tracking shows which features, prompts, and users drive cost, so you can optimize the expensive 5% instead of guessing.成本控制。按请求的 Token 和支出跟踪功能可以显示哪些功能、提示词和用户驱动了成本，这样你就可以优化那 5% 的昂贵部分，而不是盲目猜测。
Confidence to ship. Evaluation gates in CI mean a prompt change or model upgrade gets tested against real cases before it reaches users. Shipping AI changes stops feeling like a coin flip.发布的信心。CI 中的评估门禁意味着提示词更改或模型升级在到达用户之前会针对真实案例进行测试。发布 AI 变更不再像掷硬币一样令人不安。
One source of truth. When AI and application telemetry live in the same place, on-call engineers, AI engineers, and product teams argue from the same data instead of three disconnected tools.单一事实来源。当 AI 和应用程序遥测数据位于同一位置时，值班工程师、AI 工程师和产品团队可以基于相同的数据进行讨论，而不是使用三个互不关联的工具。
What to look for in an AI observability platform#选择 AI 可观测性平台时应关注什么
The category has crowded fast, and a lot of tools look similar on a feature grid. These are the capabilities that separate a platform you'll still be using at scale.该领域发展迅速，许多工具在功能列表上看起来很相似。以下是区分一个能在规模化场景下持续使用的平台的关键能力。
-
AI-native tracing. Purpose-built capture for LLM calls, agent runs, tool calls, and retrieval steps, with token counts, cost, latency, and model parameters attached to every span. Generic APM tracing bolted onto AI workloads leaves most of this out.AI 原生追踪。专为 LLM 调用、代理运行、工具调用和检索步骤构建的捕获功能，每个跨度（Span）都附带 Token 计数、成本、延迟和模型参数。附加在 AI 工作负载上的通用 APM 追踪通常会遗漏大部分此类信息。
-
Integrated evaluation. A built-in path from production traces to datasets to scored evaluations, so the observe-evaluate-improve loop lives in one platform rather than a separate eval tool you have to wire up.集成评估。从生产追踪到数据集再到评分评估的内置路径，使“观察-评估-改进”循环存在于一个平台中，而不是你需要额外连接的独立评估工具。
-
Full-stack depth. The ability to follow one request from the HTTP entry point through the agent, the model calls, the database queries, and the validation layer, in a single trace. AI bugs frequently live outside the model, and a platform that can't see the rest of the stack will send you chasing the wrong layer.全栈深度。在单个追踪中跟踪从 HTTP 入口点经过代理、模型调用、数据库查询到验证层的完整请求的能力。AI 错误经常存在于模型之外，如果平台无法看到堆栈的其余部分，它会让你在错误的层级上浪费时间。
-
Open standards, no lock-in. OpenTelemetry-native ingestion matters because it means your instrumentation is portable. If you decide to leave, your data model goes with you. Proprietary SDKs that only talk to one backend are a long-term liability.开放标准，无厂商锁定。OpenTelemetry 原生摄取非常重要，因为它意味着你的仪表化是可移植的。如果你决定离开，你的数据模型可以随你而去。只与一个后端通信的专有 SDK 是长期的负债。
-
Polyglot coverage. Your stack is not one language. Look for first-class SDKs across the languages you actually run, plus standards-based ingestion for everything else.多语言覆盖。你的技术栈不仅仅使用一种语言。寻找在你实际运行的语言中提供一流 SDK 的平台，并为其他一切提供基于标准的摄取方式。
-
Queryable data. The freedom to query raw trace data with a language you already know, rather than clicking through a fixed UI, is the difference between answering a novel question in 30 seconds and hunting through dashboards.可查询的数据。能够使用你已经熟悉的语言查询原始追踪数据，而不是点击固定的 UI，这是在 30 秒内回答一个新问题与在仪表盘中苦苦搜寻的区别。
-
Predictable pricing. Look for transparent published pricing, cost that tracks the data you send, and spend cap control rather than an open-ended bill.可预测的定价。寻找透明的公开定价、与你发送的数据量挂钩的成本，以及支出上限控制，而不是无底洞式的账单。
-
Production scale. Confirm the platform stays fast at your real trace volume. Query speed and ingestion can behave very differently between a demo and millions of daily spans, so test against your own volume before you build on it.生产规模。确认平台在你的实际追踪量下依然保持快速。查询速度和摄取能力在演示环境和每天数百万个跨度之间表现可能大相径庭，因此在构建之前请针对你自己的数据量进行测试。
Top AI observability platforms in 2026#2026 年顶级 AI 可观测性平台
The market splits roughly into eval-first platforms, LLM-only tracing tools, open-source self-hosted options, and full-stack platforms. Here is how the leading choices compare for teams running real production workloads.市场大致分为评估优先平台、仅限 LLM 的追踪工具、开源自托管选项和全栈平台。以下是领先选择在处理真实生产工作负载时的对比。
1. Pydantic Logfire: best overall for production AI applications#1. Pydantic Logfire：生产 AI 应用程序的整体最佳选择
Pydantic Logfire is an AI-native observability platform built by the team behind Pydantic, the validation library used in a large share of the world's Python AI stacks, and Pydantic AI, the agent framework.Pydantic Logfire 是一个 AI 原生可观测性平台，由 Pydantic（全球大量 Python AI 栈使用的验证库）和 Pydantic AI（代理框架）背后的团队构建。
Logfire leads on the capabilities that define AI observability. It captures every LLM call, agent run, and tool invocation with tokens, cost, latency, and model parameters attached. Conversation panels reconstruct multi-turn exchanges, tool-call inspection shows exactly what an agent did and why, and evals is wired directly into the same platform, so production traces become evaluation cases without leaving the tool. Pydantic AI Gateway, managed through Logfire, adds multi-provider model routing, cost limits, and failover. This is the observe-evaluate-improve loop most competitors stitch together from separate products, delivered in one place.Logfire 在定义 AI 可观测性的能力方面处于领先地位。它捕获每个 LLM 调用、代理运行和工具调用，并附带 Token、成本、延迟和模型参数。对话面板可以重构多轮交互，工具调用检查可以准确显示代理做了什么以及原因，评估功能直接集成在同一个平台中，因此生产追踪无需离开工具即可成为评估案例。通过 Logfire 管理的 Pydantic AI Gateway 增加了多提供商模型路由、成本限制和故障转移功能。这就是大多数竞争对手通过多个独立产品拼凑而成的“观察-评估-改进”循环，现在在一个地方即可实现。
Full-stack depth is where Logfire pulls away from the field. Most AI observability tools trace the model call and stop there. Logfire follows a single request from the HTTP route through the agent, the model and tool calls, the database queries, and the validation layer, all in one trace. When an answer goes wrong because a retrieval query timed out or an API got rate-limited, Logfire shows the real cause instead of pointing at the model. AI-only platforms structurally cannot see that, which makes whole-stack tracing a genuine differentiator rather than a checkbox.全栈深度是 Logfire 区别于其他产品的关键。大多数 AI 可观测性工具只追踪模型调用就停止了。Logfire 在一个追踪中跟踪从 HTTP 路由经过代理、模型和工具调用、数据库查询到验证层的单个请求。当答案出错是因为检索查询超时或 API 被限流时，Logfire 会显示真正的原因，而不是指向模型。仅限 AI 的平台在结构上无法看到这一点，这使得全栈追踪成为一个真正的差异化优势，而不仅仅是一个复选框。
It is built for polyglot teams. Logfire is OpenTelemetry-native from day one, so any language that speaks OTel can send data to it, and it ships first-class SDKs for Python, TypeScript, and Rust. Integrations cover the frameworks teams actually use, including Pydantic AI, FastAPI, LangChain, LlamaIndex, the Vercel AI SDK, and the OpenAI and Anthropic SDKs. Trace data is queryable with standard PostgreSQL-compatible SQL, and an MCP server lets AI coding assistants query your production traces directly, so you can ask an agent why something broke and have it read the evidence.它是为多语言团队构建的。Logfire 从第一天起就是 OpenTelemetry 原生的，因此任何支持 OTel 的语言都可以向其发送数据，并且它为 Python、TypeScript 和 Rust 提供了一流的 SDK。集成涵盖了团队实际使用的框架，包括 Pydantic AI、FastAPI、LangChain、LlamaIndex、Vercel AI SDK 以及 OpenAI 和 Anthropic SDK。追踪数据可以使用标准的 PostgreSQL 兼容 SQL 进行查询，MCP 服务器允许 AI 编码助手直接查询你的生产追踪，因此你可以询问代理某事为何中断，并让它读取证据。
Pricing is published. The free Personal plan includes 10 million spans a month, perpetually, with no credit card. Paid plans start at $49 a month, additional usage is a flat $2 per million spans, and every paid plan includes a configurable spend cap. Pydantic's pricing comparison puts Logfire at roughly 8x cheaper than Arize AX, 27x cheaper than Langfuse, and 40x cheaper than LangSmith at 5 users and 50 million spans a month. Enterprise plans add self-hosting, SSO, custom retention, and SLAs.定价是公开的。免费的个人计划永久包含每月 1000 万个跨度，无需信用卡。付费计划起价为每月 49 美元，额外使用量为每百万个跨度 2 美元的固定费用，且每个付费计划都包含可配置的支出上限。Pydantic 的定价对比显示，在 5 个用户和每月 5000 万个跨度的情况下，Logfire 比 Arize AX 便宜约 8 倍，比 Langfuse 便宜 27 倍，比 LangSmith 便宜 40 倍。企业计划增加了自托管、SSO、自定义保留期和 SLA。
Best for: engineering teams shipping production AI applications who want AI-native depth and full-stack visibility in one platform, on open standards, with transparent, predictable pricing.最适合：希望在同一个平台上获得 AI 原生深度和全栈可见性，且基于开放标准、定价透明可预测的生产 AI 应用程序工程团队。
2. Langfuse: best open-source self-hosted option#2. Langfuse：最佳开源自托管选项
Langfuse is an MIT-licensed open-source platform focused on LLM tracing, prompt management, evaluation, and dataset workflows. It self-hosts with no usage limits, which makes it a common pick for teams with strict data-residency requirements or a hard preference for owning their infrastructure. ClickHouse acquired Langfuse in January 2026; both companies committed to keeping it MIT-licensed and self-hostable, though teams evaluating it should still weigh the roadmap uncertainty that follows any acquisition. Cloud plans start around $29 a month.Langfuse 是一个 MIT 许可的开源平台，专注于 LLM 追踪、提示词管理、评估和数据集工作流。它支持自托管且没有使用限制，这使其成为对数据驻留有严格要求或倾向于拥有基础设施的团队的常见选择。ClickHouse 于 2026 年 1 月收购了 Langfuse；两家公司都承诺保持其 MIT 许可和自托管能力，尽管评估它的团队仍应权衡收购后可能带来的路线图不确定性。云计划起价约为每月 29 美元。
Best for: teams that want a self-hosted, open-source LLM tracing and prompt platform and are comfortable running their own infrastructure.最适合：希望拥有自托管、开源 LLM 追踪和提示词平台，并且乐于运行自己基础设施的团队。
3. LangSmith: best for LangChain and LangGraph stacks#3. LangSmith：LangChain 和 LangGraph 栈的最佳选择
LangSmith is the observability and evaluation layer from the LangChain team, with the tightest integration into LangChain and LangGraph. LangGraph Studio is a strong agent development environment if you live in that ecosystem. The 2026 releases added AI-assisted trace debugging and automatic behavior clustering. It is closed source, self-hosting is enterprise-only, and pricing combines per-seat fees with per-trace charges, which adds up as teams and volume grow.LangSmith 是来自 LangChain 团队的可观测性和评估层，与 LangChain 和 LangGraph 的集成最为紧密。如果你生活在那个生态系统中，LangGraph Studio 是一个强大的代理开发环境。2026 年的版本增加了 AI 辅助追踪调试和自动行为聚类。它是闭源的，自托管仅限企业版，定价结合了按席位收费和按追踪收费，随着团队和规模的增长，成本会累积。
Best for: teams committed to the LangChain and LangGraph ecosystem.最适合：致力于 LangChain 和 LangGraph 生态系统的团队。
4. Arize AI (Phoenix and Arize AX): best for evaluation-heavy ML and AI teams#4. Arize AI (Phoenix 和 Arize AX)：评估密集型 ML 和 AI 团队的最佳选择
Arize AI ships two products that share branding and trace shape but differ in license, features, and price. Phoenix is the open-source, OpenTelemetry-native tool under Elastic License 2.0, self-hostable and notebook-friendly, and a popular local-first entry point for development and evaluation. Arize AX is the commercial SaaS layered on top, adding alerts, online evals, RBAC, agent copilots, and enterprise compliance for production scale, with a Pro tier from $50 a month. Both lean on Arize's classical ML monitoring heritage, with real strength in a library of more than 50 research-backed evaluation metrics, drift detection, and embedding analysis. One thing to plan for: moving from Phoenix to AX is a new contract rather than a tier upgrade, so treat it as a procurement decision rather than an in-product step.Arize AI 推出了两款共享品牌和追踪形态但许可、功能和价格不同的产品。Phoenix 是基于 Elastic License 2.0 的开源、OpenTelemetry 原生工具，可自托管且对 Notebook 友好，是开发和评估中流行的本地优先入口点。Arize AX 是其上的商业 SaaS，增加了警报、在线评估、RBAC、代理副驾驶和企业合规性以满足生产规模，Pro 层起价为每月 50 美元。两者都依赖于 Arize 的经典 ML 监控传统，在拥有超过 50 个研究支持的评估指标库、漂移检测和嵌入分析方面具有真正的实力。需要计划的一点是：从 Phoenix 迁移到 AX 是签署新合同，而不是层级升级，因此请将其视为采购决策，而不是产品内步骤。
Best for: evaluation-heavy teams, especially those already running ML models alongside LLMs.最适合：评估密集型团队，特别是那些已经在 LLM 旁边运行 ML 模型的团队。
5. Braintrust: best for eval-driven CI/CD workflows#5. Braintrust：评估驱动的 CI/CD 工作流的最佳选择
Braintrust is an evaluation-first platform with tracing built around its eval workflow. Its strength is the trace-to-test pipeline and CI/CD gates that block a deploy when quality regresses, plus Loop, which turns natural-language descriptions into custom scorers. It runs on Brainstore, a data store Braintrust built specifically for AI workloads to keep queries fast across millions of traces. The free tier includes 1 million spans a month and 10,000 eval runs.Braintrust 是一个评估优先的平台，其追踪功能围绕评估工作流构建。它的优势在于追踪到测试的流水线和 CI/CD 门禁，当质量回归时会阻止部署，此外还有 Loop，它能将自然语言描述转化为自定义评分器。它运行在 Brainstore 上，这是 Braintrust 专门为 AI 工作负载构建的数据存储，旨在保持数百万个追踪的查询速度。免费层包含每月 100 万个跨度和 10,000 次评估运行。
Best for: teams whose primary bottleneck is regression testing and eval-gated deployment.最适合：主要瓶颈在于回归测试和评估门禁部署的团队。
6. Datadog LLM Observability: best for existing Datadog customers#6. Datadog LLM Observability：现有 Datadog 客户的最佳选择
Datadog's LLM Observability is an add-on to its established APM platform. For teams already standardized on Datadog, it adds AI tracing with zero new vendor relationship. The tradeoff is cost: AI observability pricing on Datadog runs well above usage-based specialists, and the LLM features are a recent extension rather than the core of the product.Datadog 的 LLM Observability 是其成熟 APM 平台的附加组件。对于已经标准化使用 Datadog 的团队，它增加了 AI 追踪功能，无需建立新的供应商关系。代价是成本：Datadog 上的 AI 可观测性定价远高于基于使用的专业工具，且 LLM 功能是最近的扩展，而非产品的核心。
Best for: enterprises already invested in Datadog who want AI tracing without adding a vendor.最适合：已经投资 Datadog 且希望在不增加供应商的情况下获得 AI 追踪功能的企业。
Other tools worth a look depending on your needs include Helicone for lightweight LLM logging, and Confident AI, Galileo, and Maxim AI in the evaluation-led workflows. Each covers a slice of the problem rather than the whole stack, so weigh them against what you need to see.其他值得根据需求考虑的工具包括用于轻量级 LLM 日志记录的 Helicone，以及在评估主导工作流中的 Confident AI、Galileo 和 Maxim AI。每个工具都只覆盖了问题的一部分，而不是整个堆栈，因此请根据你需要观察的内容进行权衡。
Comparison at a glance#概览对比
| Platform | Category strength | Open standards | Pricing model | Free tier |
|---|---|---|---|---|
| Pydantic Logfire | AI-native + full-stack in one trace | OTel-native; SQL-queryable; MCP | Tiered + usage, $2/M spans | 10M spans/month, perpetual |
| Langfuse | Open-source LLM tracing + prompts | OTel-compatible | Self-host free; cloud from ~$29/mo | Self-hosted; basic cloud tier |
| LangSmith | LangChain/LangGraph integration | Partial | Per-seat + per-trace | 5,000 traces/month |
| Arize AI (Phoenix / AX) | Eval metrics + ML monitoring | OTel-native (OpenInference) | Phoenix free; AX from $50/mo | Phoenix (OSS); AX free tier |
| Braintrust | Eval-first + CI/CD gates | Proprietary | Tiered + usage | 1M spans/month, 10K evals |
| Datadog | Full APM for existing customers | OTel-compatible | Enterprise, high $/span | Limited trial |
How to choose#如何选择
If your AI features are part of an application, with a backend, a database, and services around the model, lead with a platform that can trace all of it, because that's where a large share of "AI bugs" actually live. If you need to own your infrastructure outright, start with the open-source options. If you live entirely inside LangGraph, the native integration is worth weighing. If you're already on Datadog and cost is no object, the add-on removes a vendor.如果你的 AI 功能是应用程序的一部分，包含后端、数据库和模型周围的服务，请优先选择能够追踪所有这些内容的平台，因为很大一部分“AI 错误”实际上就存在于这些地方。如果你需要完全拥有自己的基础设施，请从开源选项开始。如果你完全生活在 LangGraph 内部，那么原生集成值得权衡。如果你已经在用 Datadog 且不差钱，那么附加组件可以减少一个供应商。
For most teams building production AI on an open, polyglot stack who want AI-native depth without giving up visibility into the rest of the application, Pydantic Logfire is the strongest starting point.对于大多数在开放、多语言栈上构建生产 AI，且希望在不牺牲对应用程序其余部分可见性的前提下获得 AI 原生深度的团队来说，Pydantic Logfire 是最强的起点。
Frequently asked questions#常见问题解答
What is an AI observability platform?什么是 AI 可观测性平台？
Software that captures the full execution of AI requests, including LLM calls, agent steps, tool calls, and retrieval, then attaches cost, latency, and quality data so you can debug, evaluate, and monitor production AI systems.捕获 AI 请求完整执行过程（包括 LLM 调用、代理步骤、工具调用和检索）的软件，并附加成本、延迟和质量数据，以便你可以调试、评估和监控生产 AI 系统。
How is AI observability different from traditional monitoring?AI 可观测性与传统监控有何不同？
Traditional monitoring answers whether a service is up and fast. AI observability answers whether an output was correct, why an agent behaved the way it did, and what each request cost. The failure modes are different, so the tooling is different.传统监控回答的是服务是否正常且快速。AI 可观测性回答的是输出是否正确、代理为何那样表现以及每个请求的成本。故障模式不同，因此工具也不同。
Do I need a separate tool for evaluation?我需要单独的评估工具吗？
Not if your observability platform includes it. Pydantic Logfire integrates Pydantic Evals so production traces become evaluation cases in the same platform, which removes a second tool from the loop.如果你的可观测性平台包含它，则不需要。Pydantic Logfire 集成了 Pydantic Evals，因此生产追踪可以在同一个平台中成为评估案例，这从循环中减少了第二个工具。
Is Pydantic Logfire only for Python?Pydantic Logfire 只支持 Python 吗？
No. Logfire is OpenTelemetry-native, so any language that emits OTel data works, and it ships first-class SDKs for Python, TypeScript, and Rust, with integrations across major AI frameworks.不。Logfire 是 OpenTelemetry 原生的，因此任何发出 OTel 数据的语言都可以工作，并且它为 Python、TypeScript 和 Rust 提供了一流的 SDK，并集成了主要的 AI 框架。
Can I self-host an AI observability platform?我可以自托管 AI 可观测性平台吗？
Yes, though it varies by tool. Open-source platforms like Langfuse and Arize Phoenix are self-hostable by design. Pydantic Logfire runs as a hosted cloud service by default, with self-hosting available on the enterprise tier alongside SSO, custom retention, and SLAs. Because Logfire is OpenTelemetry-native, your instrumentation stays portable either way, so the data model is yours regardless of where it runs.可以，尽管因工具而异。像 Langfuse 和 Arize Phoenix 这样的开源平台在设计上就是可自托管的。Pydantic Logfire 默认作为托管云服务运行，企业版提供自托管选项，并附带 SSO、自定义保留期和 SLA。由于 Logfire 是 OpenTelemetry 原生的，你的仪表化无论如何都是可移植的，因此无论它在哪里运行，数据模型都属于你。
What does AI observability cost?AI 可观测性的成本是多少？
It varies by model. Some platforms charge mainly per seat, some per trace or event, and some by data volume. Logfire's pricing is published in full: a free tier of 10 million spans a month, plans from $49 a month, a flat $2 per million spans for additional usage, and a configurable spend cap on every paid plan so costs stay predictable.因模式而异。有些平台主要按席位收费，有些按追踪或事件收费，有些按数据量收费。Logfire 的定价已完全公开：每月 1000 万个跨度的免费层，每月 49 美元起的计划，额外使用量为每百万个跨度 2 美元的固定费用，且每个付费计划都有可配置的支出上限，因此成本保持可预测。
Try Pydantic Logfire free#免费试用 Pydantic Logfire
You can have AI-native and full-stack traces flowing in a few minutes. The free Personal plan includes 10 million spans a month. Instrument your app with the SDK for your language, point it at Logfire, and understand what your AI is doing while getting measureably better.你可以在几分钟内让 AI 原生和全栈追踪运行起来。免费的个人计划包含每月 1000 万个跨度。使用你所用语言的 SDK 对应用程序进行仪表化，将其指向 Logfire，了解你的 AI 正在做什么，同时获得可衡量的改进。
Start free with Pydantic Logfire →免费开始使用 Pydantic Logfire →
AI is still just engineering.AI 仍然只是工程学。
