{"source":"CiteHawk AI Index","record":"LLM observability tools — 2026-11","url":"https://www.citehawk.com/leaderboards/editions/2026-11/llm-observability-tools","immutable":true,"snapshotId":"954cf034-a631-4f9f-ac14-ac7fb341ffb6","capturedAt":"2026-11-01T05:15:16.431+00:00","contentHash":"1fe906d310f375c7e845f9c922e1980a649215b9421d38dff3cc3868997bb1dd","contentHashSpec":"sha256-v1: hex SHA-256 of the UTF-8 bytes of the compact JSON array (no whitespace, non-ASCII characters unescaped, as JavaScript JSON.stringify emits) of [provider, run, model, text] tuples, one per captured answer, sorted by provider then run","registryRulesHash":"c45953a641fc763b2597a6a3f8837471f65b7550827fe50008bb0e5221fca519","registryRulesHashSpec":"sha256-v1: hex SHA-256 of the UTF-8 bytes of the compact JSON object {version, aliases, notInCategory, categoryAllow, serviceSlugRe} where aliases is the array of [foldedKey, canonicalName, pinnedDomain, categoryScope] tuples sorted by key, carrying a fifth regionScope element only on the entries that have one, notInCategory and categoryAllow are arrays of [categorySlug, domains] tuples sorted by slug, and every domain/scope list is itself sorted, carrying a trailing notInCategoryByRegion element, an array of [categorySlug, region, domains] triples sorted by slug then region, only when at least one region-scoped deny entry exists, and after it a trailing notInCategoryByName element, an array of [categorySlug, regionScope, names] triples sorted by slug then scope (regionScope \"*\" for every board region, names as published), only when at least one name-scoped deny entry exists","region":"global","prompt":"What are the best LLM observability tools? Recommend the top brands or products that people actually use.","providers":["openai","claude","gemini","perplexity","deepseek","grok","bing_copilot","google_aio","google_ai_mode"],"models":{"grok":"grok-4.3","claude":"claude-sonnet-5-5","gemini":"gemini-3.6-flash","openai":"gpt-6-sol","deepseek":"deepseek-flash","google_aio":"google_aio","perplexity":"sonar","bing_copilot":"bing_copilot","google_ai_mode":"google_ai_mode"},"runsPerProvider":1,"totalCalls":18,"ranking":[{"rank":1,"brand":"Langfuse","domain":"langfuse.com","entityId":"77d9f890-a54a-42c0-b2b7-5985c7ce7fdd","score":59.4,"mentions":9,"recommendedBy":["openai","claude","gemini","perplexity","deepseek","grok","bing_copilot","google_aio","google_ai_mode"],"averagePositionByProvider":{"grok":1,"claude":1,"gemini":1,"openai":1,"deepseek":2,"google_aio":1,"perplexity":1,"bing_copilot":1,"google_ai_mode":1}},{"rank":2,"brand":"LangSmith","domain":"langsmith.com","entityId":"aa575c3a-57d1-4e5b-8739-d82ada06817f","score":52.6,"mentions":9,"recommendedBy":["openai","claude","perplexity","deepseek","grok","bing_copilot","google_ai_mode","gemini","google_aio"],"averagePositionByProvider":{"grok":2,"claude":2,"gemini":4,"openai":2,"deepseek":1,"google_aio":2,"perplexity":2,"bing_copilot":3,"google_ai_mode":2}},{"rank":3,"brand":"Arize Phoenix","domain":"arize.com","entityId":"c7f35957-dbee-485d-9b37-d1614e26123a","score":50.9,"mentions":9,"recommendedBy":["openai","claude","gemini","perplexity","deepseek","grok","bing_copilot","google_aio","google_ai_mode"],"averagePositionByProvider":{"grok":3,"claude":3,"gemini":2,"openai":3,"deepseek":3,"google_aio":3,"perplexity":3,"bing_copilot":4,"google_ai_mode":3}},{"rank":4,"brand":"Datadog","domain":"datadog.com","entityId":"c4519313-70a4-44f0-b4b9-dc56197cbdc7","score":48,"mentions":9,"recommendedBy":["openai","claude","gemini","perplexity","deepseek","grok","bing_copilot","google_aio","google_ai_mode"],"averagePositionByProvider":{"grok":8,"claude":6,"gemini":10,"openai":5,"deepseek":7,"google_aio":5,"perplexity":6,"bing_copilot":11,"google_ai_mode":6}},{"rank":5,"brand":"Helicone","domain":"helicone.ai","entityId":"96528457-385d-42b9-80a3-8f38a92e6853","score":47.4,"mentions":8,"recommendedBy":["claude","gemini","perplexity","deepseek","grok","bing_copilot","google_aio","google_ai_mode"],"averagePositionByProvider":{"grok":4,"claude":5,"gemini":6,"deepseek":6,"google_aio":4,"perplexity":5,"bing_copilot":2,"google_ai_mode":4}},{"rank":6,"brand":"Braintrust","domain":"braintrust.com","entityId":"3ca30606-f813-4ab2-b598-6ca119e7ea29","score":47.3,"mentions":8,"recommendedBy":["openai","claude","gemini","perplexity","deepseek","grok","bing_copilot","google_ai_mode"],"averagePositionByProvider":{"grok":5,"claude":4,"gemini":5,"openai":4,"deepseek":4,"perplexity":4,"bing_copilot":7,"google_ai_mode":5}},{"rank":7,"brand":"MLflow","domain":"mlflow.org","entityId":"c94cb53e-a0e6-481c-92eb-1437f11ea9c0","score":25.9,"mentions":4,"recommendedBy":["openai","claude","perplexity","bing_copilot"],"averagePositionByProvider":{"claude":9,"openai":7,"perplexity":7,"bing_copilot":5}},{"rank":8,"brand":"Galileo","domain":null,"entityId":"22e17bd2-bc2e-4a73-a0a4-705263ba4dab","score":25.3,"mentions":4,"recommendedBy":["claude","perplexity","deepseek","gemini"],"averagePositionByProvider":{"claude":13,"gemini":9,"deepseek":8,"perplexity":9}},{"rank":9,"brand":"Weights & Biases","domain":"wandb.com","entityId":"4e5eee67-a06d-4c04-9313-8cb0fc42b2dd","score":20.3,"mentions":3,"recommendedBy":["claude","deepseek","grok"],"averagePositionByProvider":{"grok":6,"claude":7,"deepseek":5}},{"rank":10,"brand":"Portkey","domain":null,"entityId":"a33f5a1b-8f2b-4c2a-9a17-e60f231eccab","score":19.6,"mentions":3,"recommendedBy":["claude","gemini","bing_copilot"],"averagePositionByProvider":{"claude":10,"gemini":7,"bing_copilot":8}},{"rank":11,"brand":"Comet Opik","domain":"comet.com","entityId":"63eb5411-0bad-462e-860a-5fdd80c37172","score":19.5,"mentions":3,"recommendedBy":["claude","deepseek","grok"],"averagePositionByProvider":{"grok":7,"claude":8,"deepseek":11}},{"rank":12,"brand":"Confident AI","domain":"confident-ai.com","entityId":"79f83ad5-0216-4638-9c87-5f937aba4a7d","score":19.4,"mentions":3,"recommendedBy":["claude","gemini","bing_copilot"],"averagePositionByProvider":{"claude":14,"gemini":8,"bing_copilot":6}},{"rank":13,"brand":"OpenLLMetry","domain":null,"entityId":"ca12ff69-e560-4593-855f-75a73e65199b","score":13.4,"mentions":2,"recommendedBy":["deepseek","bing_copilot"],"averagePositionByProvider":{"deepseek":10,"bing_copilot":9}},{"rank":14,"brand":"Traceloop","domain":null,"entityId":"c77975b3-b738-437f-b13d-861c9a3832a2","score":13.4,"mentions":2,"recommendedBy":["deepseek","bing_copilot"],"averagePositionByProvider":{"deepseek":9,"bing_copilot":10}},{"rank":15,"brand":"AgentOps","domain":null,"entityId":"1c35115c-7ea1-4c59-9858-254a32ae70c0","score":13.1,"mentions":2,"recommendedBy":["claude","deepseek"],"averagePositionByProvider":{"claude":11,"deepseek":14}}],"answers":[{"provider":"bing_copilot","run":1,"model":"bing_copilot","capturedAt":"2026-11-01T05:15:16.438Z","text":"The most widely used LLM observability tools in 2026 are Langfuse, Helicone, LangSmith, Arize Phoenix/AX, MLflow, PromptLayer, Confident AI, and Braintrust, each dominating a different part of the tracing/evaluation ecosystem.\n\n✅ The tools people actually use most (consensus across sources)\n\nAcross the supplied results, five tools consistently appear as top picks: Langfuse, Helicone, LangSmith, Arize, and MLflow. They are repeatedly listed as leaders in open-source adoption, enterprise deployments, or developer workflows.\n\nBelow is a structured breakdown of the tools that matter most, grouped by what teams use them for.\n\n📊 1. Core LLM Observability Platforms (Tracing + Prompts + Evals)\n\nThese are the “default” choices for most engineering teams.\n\nLangfuse — Most popular open-source platform\n\nWidely cited as the best overall and the most adopted MIT‑licensed tool. Best for: full-stack tracing, prompt versioning, datasets, evaluations. Why people use it: self-hosting, strong LangChain/LlamaIndex integrations, large community.\n\nLangSmith — Best for LangChain/LangGraph users\n\nFirst-party tracing for LangChain; deep agent visibility. Best for: agent debugging, multi-step workflows, evaluation pipelines. Why people use it: unmatched fidelity for LangChain-based apps.\n\nArize Phoenix / Arize AX — Best for RAG debugging\n\nStrong retrieval observability, embedding drift, hallucination detection. Best for: enterprise RAG pipelines and semantic drift monitoring. Why people use it: excellent retrieval-level insights and drift analysis.\n\nMLflow — Best open-source end-to-end agent observability\n\nRanked #1 in MLflow’s own guide and highlighted for deep agent tracing and evaluation. Best for: teams already using MLflow or needing open-source CI/CD evaluation. Why people use it: Apache 2.0 license, strong evaluation and prompt versioning.\n\n🧪 2. Evaluation-First Platforms (Quality scoring + datasets)\n\nThese tools focus on scoring outputs, detecting hallucinations, and managing eval datasets.\n\nConfident AI — Best evaluation-first observability\n\nRanked #1 in Confident AI’s comparison; focuses on quality scoring and drift. Best for: teams needing production scoring, drift alerts, and dataset curation. Why people use it: 50+ metrics, cross-functional workflows, strong alerting.\n\nBraintrust — Best for prompt iteration and eval-driven development\n\nHighlighted as strongest for eval-first workflows. Best for: dataset-based evals, prompt experiments, regression testing. Why people use it: powerful scoring and experiment management.\n\n🔌 3. Lightweight API Observability & Gateways (Cost + latency + logs)\n\nThese tools are used when teams want fast setup and cost visibility.\n\nHelicone — Most popular proxy-based observability\n\nZero-code-change instrumentation; widely recommended. Best for: cost tracking, latency, caching, multi-provider routing. Why people use it: simplest setup (change base URL), strong cost dashboards.\n\nPortkey — Best multi-provider routing + fallback logic\n\nAI gateway with observability and routing. Best for: multi-model routing, failover, caching. Why people use it: unified logging across providers.\n\n🧱 4. Open-Source Instrumentation Layers (OTel-native)\n\nThese tools provide the underlying tracing instrumentation.\n\nOpenLLMetry / Traceloop\n\nOTel-native instrumentation used across many platforms. Best for: teams building custom observability stacks. Why people use it: vendor-neutral, integrates with 20+ backends.\n\n🧩 5. APM Platforms Adding LLM Features\n\nIf a team already uses these tools, they often start here.\n\nDatadog, Grafana Cloud, Honeycomb, New Relic\n\nAll now support LLM spans and GenAI semantic conventions. Best for: teams with existing APM infrastructure. Why people use them: unified infra + AI telemetry.\n\n🥇 Summary Table: Best Tools by Use Case\n\nUse Case | Best Tool | Why\nFull-stack tracing + evals | Langfuse | MIT, self-hostable, broad adoption 3\nLangChain/LangGraph agents | LangSmith | Deep agent tracing 2\nRAG debugging | Arize Phoenix/AX | Embedding drift + retrieval insights 3\nEvaluation-first workflows | Confident AI | 50+ metrics, drift alerts 1\nPrompt iteration + scoring | Braintrust | Dataset-based evals 3\nFastest setup (proxy) | Helicone | Zero-code-change 3\nMulti-provider routing | Portkey | Gateway + fallback logic 2\nOpen-source OTel instrumentation | OpenLLMetry / Traceloop | Vendor-neutral 2\n8 Sources8\n\n✅ Practical recommendation\n\nIf you want a single “best overall” tool: Pick Langfuse — it appears in nearly every list, is MIT-licensed, self-hostable, and widely adopted.\n\nIf you want the simplest setup: Pick Helicone — change your API base URL and you’re done.\n\nIf you need deep agent debugging: Pick LangSmith — especially if you use LangChain.\n\nIf you need evaluation-first quality monitoring: Pick Confident AI — strongest scoring and drift detection.\n\nWould you like a recommendation tailored to your stack (LangChain, RAG, multi-agent, or multi-provider)?","extractedBrands":[{"name":"Langfuse","domain":"langfuse.com","position":1},{"name":"Helicone","domain":"helicone.ai","position":2},{"name":"LangSmith","domain":"langsmith.com","position":3},{"name":"Arize Phoenix","domain":"arize.com","position":4},{"name":"MLflow","domain":"mlflow.org","position":5},{"name":"Confident AI","domain":"confident-ai.com","position":6},{"name":"Braintrust","domain":"braintrust.com","position":7},{"name":"Portkey","domain":null,"position":8},{"name":"OpenLLMetry","domain":null,"position":9},{"name":"Traceloop","domain":null,"position":10},{"name":"Datadog","domain":"datadog.com","position":11},{"name":"Grafana Cloud","domain":"grafana.com","position":12},{"name":"Honeycomb","domain":"honeycomb.io","position":13},{"name":"New Relic","domain":"newrelic.com","position":14}]},{"provider":"claude","run":1,"model":"claude-sonnet-5-5","capturedAt":"2026-11-01T05:15:16.438Z","text":"# Top LLM observability tools\n\nMost \"best of\" lists are written by vendors who rank themselves first. I've kept to the tools that show up across many independent lists.\n\n## The core picks\n\n1. **Langfuse** is the most common default for open-source and self-hosted setups. It's MIT-licensed, covers tracing, evaluation and prompt management, and is free to self-host, with cloud plans from $29/mo.\n\n2. **LangSmith** (from LangChain) is best if you build on LangChain or LangGraph. It has the deepest LangChain integration, a free tier of 5k traces/month, and a Plus plan at $39/month.\n\n3. **Arize Phoenix / Arize AX** is built on OpenTelemetry, so it avoids vendor lock-in. Phoenix is open source and OTEL-based. Arize AI as a company targets enterprise ML and LLM monitoring in high-volume production environments.\n\n4. **Braintrust** is an evaluation-first platform. It's suited to CI/CD regression gating, with a free tier and Pro at $249/mo.\n\n5. **Helicone** is the lightest way to get started. Gateway tools like Helicone and Portkey need just a URL change or header to start logging. It's a good fit if you mainly want request logging and cost tracking.\n\n## If you already have an APM\n\n- **Datadog LLM Observability** is the obvious choice if you're already a Datadog customer. It unifies APM and LLM tracing, though AI quality is secondary to infrastructure monitoring. New Relic and Dynatrace offer similar add-ons.\n\n## Other tools worth knowing\n\n- **Weights & Biases Weave** suits teams already using Weights & Biases.\n- **Comet Opik** is an open-source option with a free tier of 25k spans/month.\n- **MLflow** is worth a look if your team already uses it for ML tracking. Several lists recommend it for that case.\n- **Portkey** is a multi-provider gateway with observability.\n- **AgentOps** and **Laminar** focus on agent-focused tracing.\n- **Galileo** and **Confident AI** emphasize evaluation and guardrails.\n\n## Quick guide\n\n| Your situation | Pick |\n|---|---|\n| Want open source, self-hosted | Langfuse (or Phoenix) |\n| Built on LangChain/LangGraph | LangSmith |\n| Want OpenTelemetry, no lock-in | Arize Phoenix |\n| Eval and regression testing is the priority | Braintrust |\n| Just need cost and request logging fast | Helicone |\n| Already on Datadog | Datadog LLM Observability |\n\n## Caveats\n- Pricing and features change often, so check the vendors' own pages.\n- The sources I found were mostly vendor blogs (Comet, Confident AI, TrueFoundry, Maxim, OpenObserve), and each tends to rank its own product first. Langfuse, LangSmith, Arize Phoenix, Helicone and Datadog appear on nearly every list regardless of who wrote it. That makes them the safer bets.\n\nIf you tell me your stack (framework, cloud, self-hosting needs, budget), I can narrow this to one or two.","extractedBrands":[{"name":"Langfuse","domain":null,"position":1},{"name":"LangSmith","domain":null,"position":2},{"name":"Arize AI","domain":"arize.com","position":3},{"name":"Braintrust","domain":null,"position":4},{"name":"Helicone","domain":null,"position":5},{"name":"Datadog","domain":"datadog.com","position":6},{"name":"Weights & Biases","domain":"wandb.com","position":7},{"name":"Comet","domain":"comet.com","position":8},{"name":"MLflow","domain":null,"position":9},{"name":"Portkey","domain":null,"position":10},{"name":"AgentOps","domain":null,"position":11},{"name":"Laminar","domain":null,"position":12},{"name":"Galileo","domain":null,"position":13},{"name":"Confident AI","domain":"confident-ai.com","position":14}]},{"provider":"deepseek","run":1,"model":"deepseek-flash","capturedAt":"2026-11-01T05:15:16.438Z","text":"“Best” depends on whether you care most about tracing, evals, prompt management, cost/latency, self-hosting, or enterprise APM. The tools people actually adopt most often cluster into a few groups.\n\n## Shortlist: most commonly adopted / recommended\n\n| Tool | Best for | Why people use it | Watchouts |\n|---|---|---|---|\n| **LangSmith** | LangChain/LangGraph teams, tracing + evals | Very common in the LangChain ecosystem; traces, datasets, evals, prompt playground | Best if you’re in that ecosystem; commercial SaaS |\n| **Langfuse** | Open-source/self-hosted observability | Strong OSS + cloud; traces, prompts, evals, sessions, cost/latency; good self-host story | More DIY than enterprise APM |\n| **Arize Phoenix** | OSS tracing + evals, OTel-friendly | Popular open-source LLM tracing/eval tool; good for RAG/embeddings; Arize AX commercial option | Some features require Arize commercial |\n| **Braintrust** | Eval-first workflows and CI | Strong datasets/evals, prompt playground, regression testing; loved by AI-native teams | Less “APM monitoring,” more eval/experimentation |\n| **Weights & Biases Weave** | Teams already using W&B | LLM traces/evals alongside experiment tracking | Better fit if W&B is already in stack |\n| **Helicone** | Proxy/gateway observability, cost/latency | Easy drop-in proxy; logs requests, cost, latency, caching; popular with indie/startup devs | Proxy model may not fit every architecture |\n| **Datadog LLM Observability** | Enterprises already on Datadog | APM-style monitoring, dashboards, alerting, org-wide adoption | Can be expensive; less eval-centric than specialist tools |\n| **Galileo** | Evals, guardrails, hallucination/quality monitoring | Enterprise-focused evaluation and monitoring | Commercial; evaluate fit carefully |\n| **Traceloop / OpenLLMetry** | Vendor-neutral instrumentation | OpenTelemetry-based LLM tracing; send data to your observability backend | You own more of the stack |\n| **Comet Opik** | Open-source evals + observability | Growing OSS option for tracing/evals | Newer than incumbents |\n| **PromptLayer** | Prompt management + observability | Prompt versioning, logs, evals for product teams | Less deep APM |\n| **Humanloop** | Prompt/eval workflows for product teams | Prompt management, evaluation, collaboration | Commercial niche |\n| **AgentOps** | Agent-specific observability | Agent traces, sessions, tool calls | Best for agent workloads, not general LLM APM |\n\n## If I had to pick a “top brands” list\n\nMost people evaluating this space will shortlist:\n\n1. **LangSmith** — default for LangChain/LangGraph users.\n2. **Langfuse** — default open-source/self-hostable option.\n3. **Arize Phoenix** — strong OSS + eval/RAG observability.\n4. **Braintrust** — best-known eval-first platform.\n5. **Weights & Biases Weave** — strong for teams already in W&B.\n6. **Helicone** — easiest proxy-based cost/latency observability.\n7. **Datadog LLM Observability** — enterprise APM choice.\n8. **Galileo** — enterprise eval/guardrails choice.\n\n## Recommendation by scenario\n\n- **You use LangChain/LangGraph:** start with **LangSmith**.\n- **You want open-source/self-hosted:** **Langfuse** or **Arize Phoenix**.\n- **You care most about evals and CI regression testing:** **Braintrust**, **Weave**, or **Opik**.\n- **You want quick cost/latency logging with minimal code:** **Helicone**.\n- **You’re an enterprise already on Datadog/New Relic/Honeycomb:** use their **LLM observability** add-ons, often alongside a specialist eval tool.\n- **You’re building agents:** look at **LangSmith**, **Langfuse**, **Arize Phoenix**, or **AgentOps**.\n- **You want vendor-neutral OpenTelemetry:** **Traceloop/OpenLLMetry** + your APM backend.\n\nPractical advice: most mature teams end up with **two layers** — a specialist LLM observability/eval tool for traces, datasets, and prompt experiments, plus their existing APM/logging stack for production alerting. Don’t pick only on tracing UI; the eval/regression loop and self-host/data-privacy requirements matter more over time.","extractedBrands":[{"name":"LangSmith","domain":null,"position":1},{"name":"Langfuse","domain":null,"position":2},{"name":"Arize Phoenix","domain":"arize.com","position":3},{"name":"Braintrust","domain":null,"position":4},{"name":"Weights & Biases Weave","domain":"wandb.com","position":5},{"name":"Helicone","domain":null,"position":6},{"name":"Datadog LLM Observability","domain":"datadog.com","position":7},{"name":"Galileo","domain":null,"position":8},{"name":"Traceloop","domain":null,"position":9},{"name":"OpenLLMetry","domain":null,"position":10},{"name":"Comet Opik","domain":null,"position":11},{"name":"PromptLayer","domain":null,"position":12},{"name":"Humanloop","domain":null,"position":13},{"name":"AgentOps","domain":null,"position":14}]},{"provider":"gemini","run":1,"model":"gemini-3.6-flash","capturedAt":"2026-11-01T05:15:16.438Z","text":"The LLM observability space has matured significantly. Traditional monitoring tools (like standard APM tools) only tell you if an API returned an HTTP status 200. However, an LLM application can return a 200 while hallucinating, failing an agentic tool call, or silently degrading in response quality. \n\nModern LLM observability platforms focus on **span-level agent tracing**, **evaluations** (LLM-as-a-judge / quality checks), **prompt versioning**, **cost/token tracking**, and **hallucination/drift detection**.\n\nBelow are the top, production-tested LLM observability tools categorized by how engineering teams actually use them:\n\n---\n\n### 1. Best Open-Source & Self-Hosted Solutions\n\nIf data privacy, compliance, or self-hosting (via Docker/Kubernetes) is a requirement, these tools lead the open-source space:\n\n*   **Langfuse** *(MIT License)*\n    *   **Why people use it:** It is widely considered the default open-source choice. It offers deep tracing, session replays, LLM evaluation pipelines, and prompt management.\n    *   **Best for:** Production engineering teams that want a full-featured, self-hostable platform without vendor lock-in.\n*   **Arize Phoenix** *(Source-Available / Elastic 2.0)*\n    *   **Why people use it:** OpenTelemetry (OTel)-native and built specifically for visualizing RAG pipelines, multi-step agent reasoning, embeddings drift, and hallucination metrics. You can launch it locally inside a Jupyter notebook or run it as a microservice.\n    *   **Best for:** AI/ML researchers, RAG applications, and teams evaluating output drift over time.\n*   **Opik (by Comet)** *(Apache 2.0)*\n    *   **Why people use it:** A rapidly growing open-source platform tailored for complex agentic workflows. It combines full span-level tracing with built-in evaluation metrics and dataset management.\n    *   **Best for:** Developers building multi-agent systems who want a 100% open-source engine under Apache 2.0.\n\n---\n\n### 2. Best SaaS & Developer-First Platforms\n\nIf you want a managed solution with deep developer integration, these are the market standards:\n\n*   **LangSmith** *(by LangChain)*\n    *   **Why people use it:** The industry giant for tracing and debugging multi-step chain execution, tool calls, and complex agents. It seamlessly visualizes execution trees and simplifies dataset generation from production traces.\n    *   **Best for:** Teams using LangChain or LangGraph, though it works with any framework.\n*   **Braintrust**\n    *   **Why people use it:** Built from the ground up for **evaluation-driven development**. It makes CI/CD regression testing for prompts and models seamless, allowing you to test prompt changes against evaluation benchmarks before deploying.\n    *   **Best for:** Enterprise product & engineering teams prioritizing automated evaluation pipelines and CI integrations.\n*   **Helicone**\n    *   **Why people use it:** Known for its low-friction \"proxy approach\". By simply changing your base URL and API key, Helicone logs requests, caches responses, tracks cost/token usage, and provides basic evaluation tools.\n    *   **Best for:** Startups and teams that want plug-and-play request logging, cost analytics, and prompt management with zero SDK boilerplate.\n*   **Portkey**\n    *   **Why people use it:** Acts as an AI Gateway combined with an observability suite. Beyond tracing and monitoring, it offers production resilience features like automatic retries, fallback routing across multiple providers, and semantic caching.\n    *   **Best for:** High-throughput production apps needing gateway features (load balancing, fallbacks) alongside observability.\n\n---\n\n### 3. Best Evaluation & Quality-First Suites\n\nIf trace logging alone isn't enough and you need deep guardrails and quality validation:\n\n*   **Confident AI (powered by DeepEval)**\n    *   **Why people use it:** Built explicitly around evaluation metrics (faithfulness, answer relevance, toxicity, hallucination). It helps turn bad production traces into continuous evaluation datasets.\n    *   **Best for:** Quality assurance and teams needing rigorous metric-based evaluations on production traffic.\n*   **Galileo AI**\n    *   **Why people use it:** Provides research-backed guardrail metrics and synthetic data generation to catch low-quality or policy-violating LLM outputs in real time.\n    *   **Best for:** Enterprise risk management, compliance, and strict hallucination prevention.\n\n---\n\n### 4. Extension for Traditional APM\n\n*   **Datadog LLM Observability / SigNoz**\n    *   **Why people use it:** Traditional APM giants have added LLM-specific telemetry layer features.\n    *   **Best for:** Enterprise SRE teams already integrated into Datadog or SigNoz who want to correlate infra metrics (CPU/RAM/DB latency) directly alongside LLM spans on a single unified dashboard.\n\n---\n\n### Summary: Which one should you pick?\n\n1. **If you need open-source / self-hosted:** Pick **Langfuse** (all-around choice) or **Arize Phoenix** (for RAG & embeddings analytics).\n2. **If you use LangChain / LangGraph:** Pick **LangSmith**.\n3. **If you focus on CI/CD evaluations and regression testing:** Pick **Braintrust**.\n4. **If you want immediate drop-in monitoring & cost management:** Pick **Helicone**.\n5. **If you need an AI gateway with fallbacks & retries:** Pick **Portkey**.","extractedBrands":[{"name":"Langfuse","domain":null,"position":1},{"name":"Arize Phoenix","domain":null,"position":2},{"name":"Opik","domain":"comet.ml","position":3},{"name":"LangSmith","domain":"langchain.com","position":4},{"name":"Braintrust","domain":null,"position":5},{"name":"Helicone","domain":null,"position":6},{"name":"Portkey","domain":null,"position":7},{"name":"Confident AI","domain":null,"position":8},{"name":"Galileo AI","domain":null,"position":9},{"name":"Datadog","domain":"datadog.com","position":10},{"name":"SigNoz","domain":"signoz.io","position":11}]},{"provider":"google_ai_mode","run":1,"model":"google_ai_mode","capturedAt":"2026-11-01T05:15:16.438Z","text":"When building production-ready AI applications, developers and engineering teams turn to specialized LLM observability platforms to track costs, latency, prompts/traces, and evaluations (hallucinations, guardrails).\n\nThe top-used tools span open-source community favorites, developer-first platforms, and enterprise APM giants.\n\nTop LLM Observability & Monitoring Platforms\n\nLangfuse The Vibe: The leading open-source (MIT license), developer-favorite all-in-one platform. Best For: Teams wanting flexibility, data privacy (self-hostable), and robust tracing without immediate vendor lock-in. Core Strengths: Detailed multi-step agent debugging, prompt management, cost tracking, and analytics. It has a very generous free tier and clear pricing. LangSmith The Vibe: Built by the creators of LangChain, though it works great outside of the LangChain ecosystem too. Best For: Deep debugging of complex chains, agents, and prompt evaluation loops. Core Strengths: Exceptional trace visualization, dataset creation from production logs, and rigorous evaluation testing before pushing updates live. Arize (Phoenix & AX) The Vibe: Heavy-duty evaluation and tracing rooted in open standards like OpenInference. Best For: Data science and engineering teams prioritizing rigorous, large-scale RAG (Retrieval-Augmented Generation) and agent evaluations. Core Strengths: Arize Phoenix is a powerful open-source notebook/self-hosted tool for evaluation, while Arize AX scales up to managed enterprise production monitoring. Arize Phoenix Helicone The Vibe: Lightweight, lightning-fast proxy/gateway setup. Best For: Rapid integration, cost optimization, and caching. Core Strengths: You can set it up in minutes just by changing your API base URL. Excellent for immediate tracking of token usage, user-level cost attribution, and prompt caching. Braintrust The Vibe: \"Eval-first\" data workbench for modern AI engineering teams. Best For: Product and engineering teams treating prompt engineering and evals as a first-class citizen. Core Strengths: Seamlessly ties offline experimentation and data logging directly to production scoring and cost tracking. Datadog (LLM Observability) The Vibe: The enterprise infrastructure choice. Best For: Companies already standardized on Datadog for traditional APM and infrastructure monitoring. Core Strengths: Correlates LLM latency and token spend directly with underlying database queries, microservices, and system logs. Includes built-in security scanners for prompt injection.\n\nQuick Selection Guide\n\nIf you want open-source & self-hostable: Go with Langfuse or Arize Phoenix. If you want a 5-minute setup via proxy: Go with Helicone. If you use LangChain/LangGraph heavily: LangSmith is a natural fit. If you are an enterprise already on Datadog: Use Datadog LLM Observability to keep everything unified.\n\nTo help narrow down the ideal choice, tell me:\n\nAre you building a RAG app, an autonomous agent, or a standard chatbot ? Do you prefer an open-source/self-hosted tool or a managed cloud service ?","extractedBrands":[{"name":"Langfuse","domain":null,"position":1},{"name":"LangSmith","domain":null,"position":2},{"name":"Arize","domain":"arize.com","position":3},{"name":"Helicone","domain":null,"position":4},{"name":"Braintrust","domain":null,"position":5},{"name":"Datadog","domain":"datadog.com","position":6}]},{"provider":"google_aio","run":1,"model":"google_aio","capturedAt":"2026-11-01T05:15:16.438Z","text":"Modern LLM observability requires specialized tracing, evaluation, and token-cost tracking rather than traditional server metrics. The top LLM observability tools that teams actually use depend on your framework and scale :\n\nTop LLM Observability Tools\n\nLangfuse Best For: Teams wanting an open-source (MIT), framework-agnostic stack with generous free or low-cost self-hosting. Key Strengths: Great trace visualization, prompt versioning/management, token and cost tracking, and unlimited user seats without per-seat pricing. LangSmith (by LangChain) Best For: Engineering teams building natively on LangChain and LangGraph. Key Strengths: Deep first-party integration for step-through debugging ( LangGraph Studio ), prompt hub, and structured annotation queues. Arize Phoenix / Arize AX Best For: High-scale enterprise deployments, RAG evaluation, and open-standard OpenTelemetry/OpenInference workflows. Key Strengths: Strong open-source tracing core ( Phoenix ) paired with a heavy-duty enterprise production platform ( Arize AX ) for model/agent drift detection. Helicone Best For: Lightweight proxy-based visibility, caching, and fast setup. Key Strengths: Sits inline as an AI gateway with minimal code changes to log latencies, costs, and token usage. Datadog Best For: Large enterprises already standardizing their entire cloud infrastructure on Datadog. Key Strengths: Unified view combining traditional APM (Application Performance Monitoring) logs with LLM-specific telemetry.\n\nQuick Comparison\n\nTool License / Model Starting Price Best For\nLangfuse Open-source (MIT) / Cloud Free or $29/mo Framework-agnostic tracing & prompt management\nLangSmith Proprietary SaaS Free Developer / $39/seat/mo Plus LangChain / LangGraph native workflows\nArize Phoenix Open-source / Enterprise (AX) Free / Custom Enterprise OpenTelemetry / RAG & Agent evaluation\nHelicone Commercial / Open-source components Free tier / Usage-based Gateway-based zero-friction logging\n\nIf you'd like to narrow this down, tell me:What framework are you using (LangChain, LlamaIndex, Vercel AI SDK, or custom)?\nDo you prefer an open-source self-hosted setup or a managed cloud SaaS ?","extractedBrands":[{"name":"Langfuse","domain":null,"position":1},{"name":"LangSmith","domain":"langchain.com","position":2},{"name":"Arize Phoenix","domain":"arize.com","position":3},{"name":"Helicone","domain":null,"position":4},{"name":"Datadog","domain":"datadog.com","position":5}]},{"provider":"grok","run":1,"model":"grok-4.3","capturedAt":"2026-11-01T05:15:16.438Z","text":"**The top LLM observability tools in 2026, based on widespread adoption and expert comparisons, are Langfuse, LangSmith, Arize Phoenix (and AX), and Helicone.** These stand out across reviews for actual usage in production LLM apps and agents, with strong support for tracing (prompts, responses, tool calls, RAG retrievals, agent steps), cost/latency monitoring, evaluations (evals/scoring for quality, hallucinations, etc.), and related features like prompt management.[[1]](https://www.lilachbullock.com/best-llm-observability-tools/)[[2]](https://posthog.com/compare/best-ai-observability-tools)[[3]](https://plavno.io/blog/llm-observability-tools)\n\nOther notable mentions that see real-world use include Braintrust (eval-focused), Weights & Biases (W&B) Weave (ML experiment integration), Comet Opik (open-source with hosted tier), Datadog LLM Observability (for existing Datadog users), and gateways or OTel-based options like Traceloop/OpenLLMetry. General APM tools (e.g., PostHog for product analytics tie-ins) are sometimes extended for LLM needs.[[4]](https://logz.io/blog/top-llm-observability-tools/)[[5]](https://dupple.com/learn/best-llm-observability-tools)\n\n### Key Recommendations\nHere are the leaders, with their strengths and ideal use cases (drawn from multiple 2026 comparisons):\n\n- **Langfuse** — Frequently cited as the top or default pick for most teams. It is an open-source (MIT core) LLM engineering platform with excellent tracing, prompt versioning/management/playground, evals (LLM-as-judge, datasets, experiments), cost tracking, and sessions/agent graphs. It supports self-hosting (Docker, K8s, etc.) for data control/privacy, is framework-agnostic (strong OpenTelemetry/OpenAI SDK/LangChain/LiteLLM integrations), and offers a generous free/cloud tier (e.g., 50k units/month). Paid starts low (~$29/month Core). Acquired by ClickHouse but core remains open-source and self-hostable. Best for: Framework-agnostic teams, self-hosting needs, or balanced tracing + prompts + evals.[[2]](https://posthog.com/compare/best-ai-observability-tools)[[5]](https://dupple.com/learn/best-llm-observability-tools)[[6]](https://toolradar.com/guides/best-llm-observability-tools)\n\n- **LangSmith** (from LangChain) — The go-to for teams already using LangChain or LangGraph. It provides deep, near-zero-setup tracing (especially agent graphs and multi-step workflows), evals/datasets/annotation queues, prompt management, and debugging/replay features. Primarily hosted (free tier: ~5k traces/month, 1 seat); paid per-seat (~$39/seat Plus) with usage-based elements; enterprise self-host options. Best for: LangChain/LangGraph-heavy stacks where native integration shines.[[2]](https://posthog.com/compare/best-ai-observability-tools)[[3]](https://plavno.io/blog/llm-observability-tools)[[5]](https://dupple.com/learn/best-llm-observability-tools)\n\n- **Arize Phoenix (OSS) / Arize AX (enterprise)** — Strong open-source/source-available option (Phoenix) focused on tracing, evaluations, RAG/embedding/retrieval quality, drift detection, and ML-style monitoring. OTel-native for portability/low lock-in; self-hostable (local-first or production). Phoenix is free/self-hosted; AX adds hosted/enterprise features (~$50/month entry). Best for: RAG-heavy apps, evaluation rigor, ML/AI engineering teams, or OTel-standardized environments.[[1]](https://www.lilachbullock.com/best-llm-observability-tools/)[[7]](https://langwatch.ai/blog/top-10-llm-observability-tools-complete-guide-for-2026)[[5]](https://dupple.com/learn/best-llm-observability-tools)\n\n- **Helicone** — AI gateway/proxy with built-in observability. It offers easy setup (often just changing the base URL or one-line proxy), request-level logging/tracing, cost analytics, latency tracking, caching, and routing. Self-host options exist; free tier available, paid usage-based. Best for: Quick integration with minimal code changes, cost/spend visibility, or gateway-style needs (multi-provider routing).[[1]](https://www.lilachbullock.com/best-llm-observability-tools/)[[4]](https://logz.io/blog/top-llm-observability-tools/)\n\n### Other Strong Contenders\n- **Braintrust**: Excels at evaluation-driven workflows, experiments, datasets, and CI/CD quality gates. Good complement or alternative when evals are the bottleneck.[[2]](https://posthog.com/compare/best-ai-observability-tools)\n- **W&B Weave**: Integrates LLM observability with existing W&B experiment tracking. Ideal for ML organizations.[[2]](https://posthog.com/compare/best-ai-observability-tools)\n- **Comet Opik**: Open-source with affordable hosted tier; solid for tracing/evals on a budget.[[5]](https://dupple.com/learn/best-llm-observability-tools)\n- **Datadog LLM Observability**: Seamless add-on for teams already in the Datadog ecosystem (correlates LLM traces with infra).[[1]](https://www.lilachbullock.com/best-llm-observability-tools/)\n\n### How to Choose\n- **Stack fit** — LangChain/LangGraph → LangSmith. Otherwise or OSS preference → Langfuse or Arize Phoenix. Quick/no-code → Helicone.\n- **Self-hosting/data control** — Prioritize Langfuse or Phoenix (strong OSS/self-host parity); LangSmith is more SaaS-first.\n- **Evals vs. tracing** — Braintrust or Arize for heavy evaluation; Langfuse/LangSmith for balanced tracing + evals.\n- **Scale/pricing** — Most use usage-based models (spans/traces/observations or seats). Free tiers suffice for prototyping; self-hosting controls long-term costs but adds ops overhead. Check current pricing on vendor sites, as it evolves.[[2]](https://posthog.com/compare/best-ai-observability-tools)\n- **Portability** — Favor OTel support (most modern tools) to avoid lock-in.\n- **Real usage signals** — These tools dominate discussions on Reddit/forums and comparison roundups due to active communities, GitHub activity (e.g., Langfuse stars), and production deployments.[[8]](https://www.reddit.com/r/LangChain/comments/1vwhvfe/i_curated_48_llm_observability_tools_langfuse/)[[9]](https://www.reddit.com/r/LocalLLaMA/comments/1rjn4wf/llm_observability_is_the_new_logging_quick/)\n\n**Start with free tiers** of 1–2 options that match your stack (e.g., Langfuse + LangSmith if relevant) to test instrumentation and dashboards. Many support OpenTelemetry for easy switching later. Needs vary by team size, compliance (self-host for regulated data), volume, and whether you need deep agent tracing, RAG quality, or just cost monitoring. For the latest, review vendor docs or recent benchmarks, as the space moves quickly with acquisitions and feature updates.","extractedBrands":[{"name":"Langfuse","domain":null,"position":1},{"name":"LangSmith","domain":null,"position":2},{"name":"Arize Phoenix","domain":null,"position":3},{"name":"Helicone","domain":null,"position":4},{"name":"Braintrust","domain":null,"position":5},{"name":"Weights & Biases","domain":"wandb.com","position":6},{"name":"Comet Opik","domain":null,"position":7},{"name":"Datadog","domain":"datadog.com","position":8},{"name":"PostHog","domain":"posthog.com","position":9},{"name":"ClickHouse","domain":"clickhouse.com","position":10},{"name":"LangChain","domain":null,"position":11},{"name":"LangGraph","domain":null,"position":12},{"name":"OpenTelemetry","domain":null,"position":13},{"name":"LiteLLM","domain":null,"position":14},{"name":"RAG","domain":null,"position":15}]},{"provider":"openai","run":1,"model":"gpt-6-sol","capturedAt":"2026-11-01T05:15:16.438Z","text":"**My shortlist is Langfuse, LangSmith, Arize Phoenix, Braintrust, Datadog, Sentry, and MLflow.** There isn’t one universal winner: the right choice depends on whether you primarily need to debug agent runs, evaluate answer quality, or monitor production operations.\n\n| Product | Best fit | Why I’d consider it |\n|---|---|---|\n| **Langfuse** | Best general-purpose starting point, especially if you want to self-host | Agent traces, cost tracking, evaluations, datasets, and prompt management in one platform. Its core features can be self-hosted. ([langfuse.com](https://langfuse.com/docs/observability/overview?trk=public_post_comment-text&utm_source=openai)) |\n| **LangSmith** | Teams using LangChain or LangGraph; also suitable for other stacks | Particularly strong workflow from detailed traces to human review, evaluations, and production monitoring. It is **not** limited to LangChain applications. ([langchain.com](https://www.langchain.com/langsmith/observability?utm_source=openai)) |\n| **Arize Phoenix** | Local or self-hosted agent and RAG debugging | Clear step-by-step traces alongside evaluations and experiments. Check its **ELv2 license** against your requirements rather than assuming “open source” means permissively licensed. ([arize.com](https://arize.com/phoenix/?utm_source=openai)) |\n| **Braintrust** | Teams that use evaluations to decide what ships | Connects production traces and quality scores to datasets and regression tests. ([braintrust.dev](https://www.braintrust.dev/learn/ai-observability/v0?utm_source=openai)) |\n| **Datadog Agent Observability** | Organizations already running Datadog | Brings agent traces, token usage, cost, alerts, and quality evaluations into the same operational view as application monitoring. ([docs.datadoghq.com](https://docs.datadoghq.com/llm_observability/?trk=article-ssr-frontend-pulse_little-text-block&utm_source=openai)) |\n| **Sentry** | Teams already using Sentry for application errors | Connects agent steps and LLM calls to application traces, with token and cost visibility. ([sentry.io](https://sentry.io/solutions/ai-observability/?utm_source=openai)) |\n| **MLflow** | Teams already invested in MLflow or wanting a self-managed option | Open-source, OpenTelemetry-compatible tracing with evaluation and production-monitoring workflows. ([mlflow.org](https://mlflow.org/docs/latest/llms/tracing/?utm_source=openai)) |\n\nThese are **products people actually report using**, not just names from comparison lists. In the 2026 Stack Overflow Developer Survey’s AI-observability responses, Sentry was reported by 26.0%, LangSmith by 17.3%, MLflow and Langfuse by 15.2% each, Datadog by 11.9%, and Arize and Braintrust by 5.8% each. Those are survey responses, **not market shares**. ([survey.stackoverflow.co](https://survey.stackoverflow.co/2026/ai/data/ai-obs?utm_source=openai))\n\n**If you want one to try first:** pick **Langfuse** for a flexible dedicated tool; **LangSmith** if you’re building on LangGraph; **Braintrust** if your main question is “did this change improve quality?”; or **Datadog/Sentry** if keeping AI and application incidents together matters most. **Comet Opik** is another credible self-hostable alternative worth testing alongside Langfuse and Phoenix. ([langfuse.com](https://langfuse.com/docs/observability/overview?trk=public_post_comment-text&utm_source=openai))\n\nOne dated caveat: I **wouldn’t start a new deployment on Helicone without checking its roadmap**. Helicone announced on **March 3, 2026** that, following its acquisition by Mintlify, its service would remain live in *maintenance mode*. ([helicone.ai](https://www.helicone.ai/blog/joining-mintlify?utm_source=openai))","extractedBrands":[{"name":"Langfuse","domain":"langfuse.com","position":1},{"name":"LangSmith","domain":null,"position":2},{"name":"Arize Phoenix","domain":"arize.com","position":3},{"name":"Braintrust","domain":"braintrust.dev","position":4},{"name":"Datadog","domain":"datadoghq.com","position":5},{"name":"Sentry","domain":"sentry.io","position":6},{"name":"MLflow","domain":"mlflow.org","position":7},{"name":"Helicone","domain":"helicone.ai","position":8},{"name":"Mintlify","domain":null,"position":9}]},{"provider":"perplexity","run":1,"model":"sonar","capturedAt":"2026-11-01T05:15:16.438Z","text":"The strongest practical shortlist is **Langfuse, LangSmith, Arize Phoenix, Braintrust, Helicone, Datadog, and MLflow**. The best choice depends less on brand popularity than on your framework, deployment requirements, and whether you need tracing, evaluation, cost monitoring, or full-stack APM.\n\n| Product | Best for | Why teams use it |\n|---|---|---|\n| **Langfuse** | Most teams; open-source or self-hosted deployments | Framework-agnostic tracing, prompt management, sessions, evaluations, and strong self-hosting support. Often the safest default for teams that want flexibility and low vendor lock-in.[1][3][5] |\n| **LangSmith** | LangChain and LangGraph applications | Particularly strong tracing and debugging for chains, agents, and graph workflows; the natural choice for teams already committed to the LangChain ecosystem.[1][5][6] |\n| **Arize Phoenix** | RAG quality, drift, and OpenTelemetry | Open-source tracing with strong retrieval and generation evaluation capabilities; a good fit for teams that want vendor-neutral instrumentation.[1][3][5] |\n| **Braintrust** | Evaluation-driven development | Combines production traces with experiments, datasets, human review, and rapid trace search; useful when evaluation is part of the development workflow rather than an afterthought.[1][4][5] |\n| **Helicone** | Fast setup, cost, and usage monitoring | Proxy-based observability makes it easy to capture requests, tokens, latency, provider usage, and spend with relatively little instrumentation.[1][5] |\n| **Datadog LLM Observability** | Enterprises already using Datadog | Best when LLM traces need to correlate with existing application, infrastructure, security, and incident-monitoring data.[4][8] |\n| **MLflow** | ML-platform teams and end-to-end lifecycle management | A strong option for teams already using MLflow for experiment tracking, model management, deployment, and GenAI evaluation.[5][6] |\n| **W&B Weave** | Teams already using Weights & Biases | Adds LLM tracing and evaluation to an established ML experimentation platform.[2][7] |\n| **Galileo** | Enterprise quality monitoring and guardrails | Focuses on automated quality evaluation, runtime monitoring, and governance, especially for regulated or high-stakes applications.[4][7] |\n| **Pydantic Logfire** | Python and Pydantic-heavy applications | Useful when application, model, and LLM traces should live in one OpenTelemetry-oriented system.[8] |\n\n## Recommended picks by situation\n\n- **Default recommendation:** **Langfuse**—particularly for a multi-framework team that wants open source, self-hosting, and broad tracing capabilities.[3][5]\n- **LangChain/LangGraph stack:** **LangSmith**.[1][5]\n- **RAG-heavy application:** **Arize Phoenix** or **Langfuse**; Phoenix has especially strong positioning around retrieval debugging and evaluation.[1][3]\n- **Agent debugging:** **LangSmith**, **Langfuse**, or **Braintrust**.\n- **Evaluation-first quality program:** **Braintrust** or **Confident AI**; these emphasize turning traces into datasets, experiments, and recurring evaluations.[1][4]\n- **Cost and provider monitoring:** **Helicone**; consider a gateway such as Portkey when routing and fallback across model providers also matter.[1][10]\n- **Existing enterprise observability platform:** Use **Datadog**, **Dynatrace**, or **New Relic** if your organization already standardizes on one of them.[2][4][10]\n- **Self-hosting and data residency:** **Langfuse** or **Arize Phoenix**.[1][3][5]\n- **Full ML lifecycle rather than only LLM tracing:** **MLflow**.[5][6]\n\n## Bottom line\n\nFor most new projects, start with **Langfuse** unless the application is deeply built around LangChain, in which case choose **LangSmith**. Add **Arize Phoenix** when RAG evaluation and OpenTelemetry matter, **Helicone** when cost visibility is the immediate priority, and **Datadog** when LLM monitoring must integrate with an existing enterprise APM environment. These products appear repeatedly across current industry shortlists and usage-oriented comparisons.[1][2][5][9]","extractedBrands":[{"name":"Langfuse","domain":null,"position":1},{"name":"LangSmith","domain":null,"position":2},{"name":"Arize Phoenix","domain":null,"position":3},{"name":"Braintrust","domain":null,"position":4},{"name":"Helicone","domain":null,"position":5},{"name":"Datadog","domain":"datadog.com","position":6},{"name":"MLflow","domain":"mlflow.org","position":7},{"name":"W&B Weave","domain":null,"position":8},{"name":"Galileo","domain":null,"position":9},{"name":"Pydantic Logfire","domain":null,"position":10}]}]}