--- type: analysis slug: guide-agent-observability-tools title: "Tracing agents: Langfuse, LangSmith, Braintrust, Phoenix" description: "Langfuse, LangSmith, Braintrust and Arize Phoenix are platforms for tracing and evaluating LLM applications and agents: they record each run as a trace of model calls, tool use, retrieval and custom steps, track tokens and cost, and score outputs with LLM-as-judge, code or human review. Langfuse and" published_at: 2026-11-05T09:00:00.000Z topics: ["tools","developer_tools"] source_urls: ["https://langfuse.com/pricing","https://langfuse.com/integrations/native/opentelemetry","https://www.langchain.com/pricing","https://docs.langchain.com/langsmith/trace-with-opentelemetry","https://www.braintrust.dev/docs/integrations/sdk-integrations/opentelemetry/send-traces-and-logs","https://www.braintrust.dev/pricing","https://arize.com/docs/phoenix","https://arize.com/pricing/","https://langfuse.com/self-hosting","https://arize.com/docs/phoenix/self-hosting","https://langfuse.com/changelog"] author: "Andrey Reshetnikov" author_profile: https://newruntime.com/owner-profile.md --- Langfuse, LangSmith, Braintrust and Arize Phoenix are platforms for tracing and evaluating LLM applications and agents: they record each run as a trace of model calls, tool use, retrieval and custom steps, track tokens and cost, and score outputs with LLM-as-judge, code or human review. Langfuse and Phoenix are open source and self-hostable (Phoenix with no feature limits); LangSmith comes from LangChain and adds agent deployment; Braintrust is a hosted eval-first platform with self-hosting on Enterprise. All four accept OpenTelemetry traces. They are for teams moving agents to production who need to debug runs, run experiments and watch quality and cost. *Field guide compiled 2026-10-23 from the official documentation, pricing pages and changelogs linked below; every line links to the page that states it.* ## What it does - **LLM Application & Agent Tracing.** Langfuse records traces and agent graphs, sessions (chats/threads), users, and token and cost usage. How: Langfuse Python or JS/TS SDK, native framework integrations, OpenTelemetry, or LiteLLM proxy logging [source](https://langfuse.com/pricing) - **OpenTelemetry (OTEL) for LLM Observability.** Send OTEL traces from any application, framework or collector to Langfuse; Langfuse recommends its SDK for Python and JS/TS. How: Configure an OTLP exporter to Langfuse per the OpenTelemetry integration page [source](https://langfuse.com/integrations/native/opentelemetry) - **Evaluation (online and offline).** Langfuse datasets, experiments via SDK and UI, custom scores, user feedback, LLM-as-judge evaluators and human annotation queues. How: Langfuse Evaluation features; Hobby includes 1 annotation queue [source](https://langfuse.com/pricing) - **LangSmith Observability & Evaluation.** Tracing, monitoring, insights, online and offline evals, dataset collection, annotation queues, Prompt Hub and Playground, bulk export. How: LangSmith SDK or integrations for LangChain, LangGraph, OpenAI Agents SDK, Claude Agent SDK, CrewAI and others [source](https://www.langchain.com/pricing) - **Trace with OpenTelemetry.** LangSmith ingests OTel traces from LangChain and non-LangChain apps and maps GenAI standard, TraceLoop, OpenInference and Logfire attributes. How: Configure OTLP endpoint and environment variables per the LangSmith OpenTelemetry guide [source](https://docs.langchain.com/langsmith/trace-with-opentelemetry) - **LangSmith Deployment.** Deploy and manage long-running agents, expose an agent as an MCP server, stream intermediate steps, scale to zero when idle. How: Plus plan includes 1 free Serverless (Small) deployment [source](https://www.langchain.com/pricing) - **Send OpenTelemetry traces and logs to Braintrust.** Braintrust accepts traces and logs through an OTLP trace exporter or a Braintrust span processor. How: Point an OTLP exporter at Braintrust or add the Braintrust span processor [source](https://www.braintrust.dev/docs/integrations/sdk-integrations/opentelemetry/send-traces-and-logs) - **Loop agent.** Braintrust's built-in AI agent that runs evaluations, generates test cases and iterates on prompts. How: Available in the Braintrust platform [source](https://www.braintrust.dev/pricing) - **Tracing in Phoenix.** Phoenix accepts traces over OpenTelemetry (OTLP) and auto-instruments LlamaIndex, LangChain, DSPy, Mastra, Vercel AI SDK, OpenAI, Bedrock, Anthropic in Python, TypeScript and Java via OpenInference. How: Install OpenInference instrumentation and export OTLP to a Phoenix instance [source](https://arize.com/docs/phoenix) - **Evals in the Phoenix UI.** Score traces and spans with LLM-based evaluators, code checks or human labels; bring evaluators from Ragas, Deepeval or Cleanlab. How: Phoenix evals, dataset evaluators and human annotations in the UI [source](https://arize.com/docs/phoenix) ## Modes - **Langfuse Cloud.** Fully managed by the Langfuse team; data region US, EU or JP. [source](https://langfuse.com/self-hosting) - **Docker Compose.** Langfuse self-hosting for local use and testing on a single VM without high availability, scaling or backups. [source](https://langfuse.com/self-hosting) - **Kubernetes (Helm).** Langfuse production self-hosting (also AWS, Azure, GCP); depends on PostgreSQL, Redis/Valkey, ClickHouse and S3/blob storage. [source](https://langfuse.com/self-hosting) - **Phoenix self-hosting.** Free to self-host with no feature limitations, can be fully air-gapped; Docker, Kubernetes, Cloud Run and others. [source](https://arize.com/docs/phoenix/self-hosting) - **LangSmith Enterprise.** Self-hosted and hybrid deployment options with custom SSO, ABAC and RBAC. [source](https://www.langchain.com/pricing) - **Braintrust Enterprise.** On-prem or hosted deployment for high-volume or privacy-sensitive data. [source](https://www.braintrust.dev/pricing) ## Pricing and limits - **Langfuse Core:** $29 / month. 100k units / month included, additional $8/100k units, 90 days data access, unlimited users [source](https://langfuse.com/pricing) - **Langfuse Pro:** $199 / month. 100k units included then $8/100k, 3 years data access, high rate limits; Teams add-on $300/mo [source](https://langfuse.com/pricing) - **Langfuse Enterprise:** $2499 / month. Audit logs, SCIM API, custom rate limits, uptime and support SLA [source](https://langfuse.com/pricing) - **LangSmith Developer:** $0 / seat per month then pay as you go. Up to 5k base traces / mo, 1 seat [source](https://www.langchain.com/pricing) - **LangSmith Plus:** $39 / seat per month then pay as you go. Up to 10k base traces / mo, unlimited seats, 1 free Serverless (Small) deployment; usage metered at $1.00 / LSU [source](https://www.langchain.com/pricing) - **Braintrust Starter:** $0 / month. $10 credits, 1 GB processed data + $4/GB, 10k scores + $2.50/1k, 14-day retention, unlimited users [source](https://www.braintrust.dev/pricing) - **Braintrust Pro:** $249 / month. $100 credits, 5 GB processed data + $3/GB, 50k scores + $1.50/1k, 30-day retention + $0.50/GB/mo [source](https://www.braintrust.dev/pricing) - **Arize AX Pro:** $50 per month. 50k spans per month, 10 GB ingestion, 30 days retention [source](https://arize.com/pricing/) ## Recent changes - 2026-10-22: Langfuse self-hosted Enterprise can run in FIPS mode on FIPS 140-3 validated modules; images rebuilt on Iron Bank hardened Red Hat UBI 9. [source](https://langfuse.com/changelog) - 2026-10-21: Langfuse can export score metadata and provenance to S3, GCS and Azure scheduled exports. [source](https://langfuse.com/changelog) - 2026-10-18: Langfuse experiments chart every score across runs and compare candidates item by item against a baseline. [source](https://langfuse.com/changelog) - 2026-10-15: Langfuse project API keys can be created with limited permissions. [source](https://langfuse.com/changelog) ## Alternatives - **Arize AX.** Arize's managed enterprise platform built on the same open standards as Phoenix, with paid tiers from $50 per month and self-hosting only on Enterprise. [source](https://arize.com/pricing/) ## Traps - Langfuse self-hosting with Docker Compose has no high availability, scaling or backups; production needs Kubernetes or a cloud deployment with PostgreSQL, ClickHouse, Redis and blob storage. [source](https://langfuse.com/self-hosting) - Some Langfuse self-hosted add-on features require a license key. [source](https://langfuse.com/self-hosting) - Langfuse Hobby API rate limits are low: 30 requests/min general API and 100 requests/day for Metrics API v2. [source](https://langfuse.com/pricing) - LangSmith free Developer plan is limited to 1 seat and 5k base traces per month before pay-as-you-go. [source](https://www.langchain.com/pricing) - Braintrust Starter keeps data only 14 days and allows 1 human review score per project. [source](https://www.braintrust.dev/pricing) Author: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json ## Continue in the graph Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`. - topic hub: [Tools](https://newruntime.com/topics/tools.md): Explore the tools topic hub. - topic hub: [Developer Tools](https://newruntime.com/topics/developer-tools.md): Explore the developer_tools topic hub. - collection: [Analysis](https://newruntime.com/analysis.json): Open the latest authorial analysis. - collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites. - whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)