Langfuse is live · free plan, paid from $29/mo · last verified 6 Nov 2026 · viability healthy
Langfuse
Open-source LLM observability and agent evals: trace, evaluate, and improve AI agents on one MIT-licensed platform
Langfuse earns its adoption because tracing, prompts, evals, and human review share one data model instead of four stitched tools. If your team already emits OpenTelemetry, the migration cost is close to zero. Watch the meter: Core is $29/mo and Pro $199/mo, each including 100k units/month, with usage past that at $8/100k units and volume discounts down to $6/100k above 50M.
Verdict generatedVerified 7d ago · liveness 89/100 · cite: rightaichoice.com/tools/langfuse
- AI engineering teams tracing multi-turn chat, coding, or workflow agents in production
- Enterprises needing self-hosting, FIPS 140-3 mode, or a US/EU/JP data region for SOC 2, ISO 27001, or HIPAA requirements
- Teams consolidating tracing, prompt management, datasets, experiments, and human review on one data model
- Organizations using IDE coding agents that want prompts and traces reachable via MCP, CLI, or SKILL.md
- Solo developers who want a zero-config logger and won't run ClickHouse, Redis, and blob storage to self-host
- Teams expecting flat-rate pricing — paid tiers include 100k units/month, then meter at $8/100k units
- Buyers who want a fully managed black box and won't manage data portability
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Langfuse if you want a zero-ops logger you never tune — the Cloud tiers meter at $8/100k units past the included 100k, and self-hosting means running ClickHouse, Redis, and blob storage yourself.
Every paid tier includes 100k units/month, then bills $8/100k units — a high-volume agent can cost multiples of the $29 or $199 sticker price.
Hobby is genuinely free with 50k units/month, so a prototype costs nothing. Core at $29/mo fits a small production team; Pro at $199/mo adds 3-year retention, unlimited annotation queues, higher rate limits, and SOC 2/ISO 27001 reports — typically the tier a company buying for compliance lands on. Enterprise at $2499/mo buys SSO, RBAC, audit logs, SCIM, and SLAs. All paid tiers include 100k units/month then meter at $8/100k, so compare total spend at your trace volume, not the sticker.
In short
Langfuse — Open-source LLM observability and agent evals: trace, evaluate, and improve AI agents on one MIT-licensed platform. Best for AI engineering teams tracing multi-turn chat, coding, or workflow agents in production, Enterprises needing self-hosting, FIPS 140-3 mode, or a US/EU/JP data region for SOC 2, ISO 27001, or HIPAA requirements, Teams consolidating tracing, prompt management, datasets, experiments, and human review on one data model. Free to start; paid plans from $29/mo.
What is Langfuse?
Langfuse — Open-source LLM observability and agent evals: trace, evaluate, and improve AI agents on one MIT-licensed platform. Best for AI engineering teams tracing multi-turn chat, coding, or workflow agents in production, Enterprises needing self-hosting, FIPS 140-3 mode, or a US/EU/JP data region for SOC 2, ISO 27001, or HIPAA requirements, Teams consolidating tracing, prompt management, datasets, experiments, and human review on one data model. Free to start; paid plans from $29/mo.
Langfuse is the open-source AI engineering platform for tracing, evaluating, and improving LLM and agent applications. It's built for engineering teams shipping agents to production, not for analysts hunting for a chart builder. Hierarchical traces capture every LLM call, tool invocation, and retrieval step, filterable by user, session, cost, latency, or custom metadata, and agentic workflows can be viewed as graphs. The platform closes one loop: observability, prompt and skill management, evaluation, experiments, and human feedback. LLM-as-a-judge evaluators, heuristic functions, user feedback, and human annotation queues score production traffic or experiment runs. Prompt management handles versioning, labels, one-click deploys and rollbacks, plus server- and client-side caching. A playground tests prompts on real production inputs and compares models side by side. Recent releases added a Langfuse AI Gateway that routes OpenAI and Anthropic calls through a base-URL swap, traces them automatically, and reports cost per gateway key and per user. October 2026 also brought Skill Management, experiment charts that show which metric moved against a baseline, org-level LLM connections shared across projects, and permission-scoped project API keys. Halo, the in-app assistant launched in October 2026, investigates production data, finds failed generations and high-latency traces, builds eval datasets, and takes actions with user approval. SKILL.md, a CLI, and an MCP server wire IDE coding agents to the same data. Ingestion is OpenTelemetry-native across Python, TypeScript, Go, Java, .NET, Ruby, PHP, and Swift with no framework lock-in, and Python SDK v5 plus JS/TS SDK v6 shipped for v4. Self-host with Docker Compose, Helm, or Terraform, including a FIPS 140-3 mode on hardened images, or use Cloud in US, EU, or JP regions. Positioning: the MIT-licensed answer to managed observability suites, with S3/blob export keeping data portable. Compared with a zero-config logger
Key features
- Hierarchical tracing of LLM calls, tool invocations, retrieval, and other API calls
- Session tracking for multi-turn chat and agentic workflows
- Agent graphs that visualize complex agentic workflows
- OpenTelemetry-native SDKs for Python, TypeScript, Go, Java, .NET, Ruby, PHP, and Swift
- Langfuse AI Gateway: route OpenAI and Anthropic calls with automatic tracing and cost per user
- LLM-as-a-judge evaluators with multi-message prompts and multi-modal inputs
- Skill Management for versioned agent skills
- Experiment charts that show which metric moved and why, against a baseline
- Human annotation queues for human-in-the-loop review and golden datasets
- Prompt versioning, labels, one-click deploys, and rollbacks
- Prompt composability with server- and client-side prompt caching
- Playground for testing prompts on production inputs and comparing models
- Halo in-app assistant that investigates traces, builds datasets and dashboards, and takes approved actions
- MCP server, CLI, and SKILL.md for connecting IDE coding agents
- Scheduled exports of traces and scores to S3, GCS, and Azure blob storage
Behind the verdict
Pick Langfuse when your agents are real and so is your bill. The feature that decides it for most teams isn't a feature at all — it's the shared data model. A trace you capture in production becomes an eval input, a dataset item, a prompt experiment, and a dashboard with no re-instrumentation. Add org-level LLM connections and permission-scoped API keys from October 2026, and a platform team can hand projects to app teams without handing over the keys to everything. The AI Gateway is the quiet upgrade. Swapping the base URL and API key routes OpenAI and Anthropic calls through Langfuse, so every call is traced without touching app code, and you get cost per user and per key. If you already run LiteLLM or OpenRouter, nothing forces you onto it. Pass when nobody owns evals. Langfuse will happily store annotation queues you never open and dashboards nobody reads. Small teams that just want to see exceptions in a log will find the setup, and especially self-hosting ClickHouse, Redis, and blob storage, heavier than the problem. The closest alternative is LangSmith for teams already deep in the LangChain ecosystem, and Arize Phoenix for a lighter OSS trace viewer. Langfuse's differentiator is breadth on the MIT license plus genuine portability through scheduled S3, GCS, and Azure exports. What bites: unit-based billing rewards verbose agents that emit many observations, so model your monthly volume before you pick a tier — 90 days of data on Core versus three years on Pro quietly changes the compliance answer too.
Limitations
- Langfuse is an observability and AI engineering platform, not a model provider — you bring your own LLM and pay that provider separately (or route calls through the Langfuse AI Gateway pointing at OpenAI or Anthropic).
- Cloud tiers are metered by units: Hobby is free with 50k units/month, 30 days data access, and 2 users; Core is $29/month and Pro is $199/month, each including 100k units/month with additional usage at $8/100k units.
- Self-hosting is supported (Docker Compose, Kubernetes/Helm, Terraform) but means operating the infrastructure yourself.
as of 2026-10-19
Researching Langfuse? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is Langfuse still active?
Langfuse is live · free plan, paid from $29/mo · last verified 6 Nov 2026 · viability healthy.
Viability score
How well maintained and how widely used is Langfuse? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: November 2026
How we score →What's new in Langfuse
Checked 5 days agoAcross the latest 10 updates: 6 feature updates, 2 launches, 1 changelog entry and 1 community discussion.
S3 bucket connections via IAM role for blob exports (Enterprise)
Blob storage exports and External Media Storage can assume an IAM role in your AWS account instead of using stored access keys. Enterprise only.
Langfuse: missing token usage now visible as automatic tokenization retires
Langfuse is retiring automatic tokenization so missing usage is visible instead of hidden behind misleading cost estimates.
How Teams in Regulated Industries Get AI Agents into Production
Guide on taking an AI agent from offline prototype to production for banks, insurers, and healthcare teams, covering evidence, compliance checks, and human oversight.
Langfuse AI Gateway enters public beta
Point apps and coding agents at Langfuse instead of OpenAI or Anthropic, issue gateway keys instead of provider keys, and see cost per user, key, and model with every call traced.
Skill Management in Langfuse now in beta
Collaborate on and version AI agent skills in Langfuse, and connect them with your agent traces. Available in beta.
Continuation cursors for expensive observations queries
Short or empty pages can include a continuation cursor so expensive observations queries can make progress across requests instead of timing out.
Langfuse alerts can be sent to PagerDuty
Receive alerts, project notifications, and prompt changes as PagerDuty events to page an on-call team from Langfuse.
Langfuse Slack app listed on the Slack Marketplace
The Langfuse Slack app is now an official Slack Marketplace app, installable in workspaces that only allow Marketplace apps.
Self-hosted Langfuse adds FIPS 140-3 mode
Government deployments can enable FIPS mode, running Langfuse cryptography on FIPS 140-3 validated modules and refusing to start otherwise. Web and worker images now build on Iron Bank's hardened Red Hat UBI 9.
Langfuse launches Halo, an in-app assistant
Halo, the assistant inside Langfuse, can investigate traces, help set up evals, build dashboards, and take action with user approval.
What do users say about Langfuse?
What people actually say about Langfuse — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
ResearchedDiscussed by users across 6 sources (Hacker News, YouTube, Product Hunt, Stack Overflow, GitHub, Lemmy) · researched Sep 9, 2026.
Weighted by the 73 posts each of 6 sources contributed.
- +Open-source and MIT-licensed, avoiding vendor lock-in with self-hosting.
- +Hierarchical traces give great depth for debugging LangChain and LangGraph.
- +Recent v4 boasts up to 165x faster real-time processing with ClickHouse.
- +LLM-as-a-judge and human annotation queues are easy to set up.
- +100+ integrations cover most frameworks and providers.
- −Learning curve for beginners; UI/UX less intuitive than some alternatives.
- −Initial setup can be complex, especially for self-hosting.
- −Documentation sometimes too high-level; concrete examples missing.
- −Native SDKs don't cover all languages, requiring extra components.
- −Some users still miss error-specific tracking features.
- • If self-hosting, you incur infrastructure costs (ClickHouse, etc.).
- • Custom retention and advanced features may require enterprise plan.
How we verified this
Every listed fact about Langfuse — pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 6 Nov 2026. Analysis is drafted with AI models and checked against our editorial standards before publishing; see how we verify.
Verification history
We have re-verified Langfuse 23 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed limitations, pricing page
- — changed description, our verdict, pricing tiers, integrations, summary, features
- — changed pricing tiers, our verdict, integrations, features, who it suits, description, summary
- — changed models, pricing page
- — changed who should skip it, pricing tiers, summary, our verdict, features, who it suits, description
- — changed pricing page, limitations
Showing the 6 most recent of 23 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does Langfuse cost in 2026?
Langfuse: Free to start; paid plans from $29/mo. Pricing last verified 6 Nov 2026.
Plans compared
Every published Langfuse tier, its price as the vendor publishes it, who it fits, and what it adds over the tier below. The 12-month cost projection follows.
| Plan | Price | Billing | Who it's for |
|---|---|---|---|
| Hobby | Free | — | A solo developer or small POC tracing a first agent, with no credit card and 2 users.Adds: Free entry point: all platform features with limits, 50k units/month, 30 days of data access, 2 users, and community support via GitHub. |
| Core | $29/mo | monthly | A small production team that has outgrown a POC and needs unlimited users and 90-day history.Adds: Adds unlimited users, 100k included units/month, 90-day data access, 4,000 requests/min ingestion, and in-app support over Hobby for $29/mo. |
| Pro | $199/mo | monthly | A scaling team that needs 3-year retention, unlimited annotation queues, higher rate limits, and compliance reports.Adds: Adds 3 years of data access, data retention management, unlimited annotation queues, high rate limits, prioritized support, and SOC 2/ISO 27001 reports with a HIPAA-ready region over Core, for $199/mo. |
| Teams (add-on) | $300/mo | monthly | — |
| Enterprise | $2,499/mo | monthly | A large organization with security review requirements — SSO, RBAC, audit logs — and procurement that bills by invoice or AWS Marketplace.Adds: Adds SSO with enforcement, fine-grained RBAC, audit logs, SCIM API, custom rate limits, uptime and support SLAs, and a named lead support engineer over Pro, for $2499/mo. |
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where Langfuse's pricing actually pencils out — and where peers do it cheaper.
Hobby is genuinely free with 50k units/month, so a prototype costs nothing. Core at $29/mo fits a small production team; Pro at $199/mo adds 3-year retention, unlimited annotation queues, higher rate limits, and SOC 2/ISO 27001 reports — typically the tier a company buying for compliance lands on. Enterprise at $2499/mo buys SSO, RBAC, audit logs, SCIM, and SLAs. All paid tiers include 100k units/month then meter at $8/100k, so compare total spend at your trace volume, not the sticker.
Has Langfuse's pricing changed?
No plan price for Langfuse has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick Langfuse?
Pick Langfuse if you are AI engineering teams tracing multi-turn chat, coding, or workflow agents in production, Enterprises needing self-hosting, FIPS 140-3 mode, or a US/EU/JP data region for SOC 2, ISO 27001, or HIPAA requirements, Teams consolidating tracing, prompt management, datasets, experiments, and human review on one data model, Organizations using IDE coding agents that want prompts and traces reachable via MCP, CLI, or SKILL.md.
Look elsewhere if you are Solo developers who want a zero-config logger and won't run ClickHouse, Redis, and blob storage to self-host, Teams expecting flat-rate pricing — paid tiers include 100k units/month, then meter at $8/100k units, Buyers who want a fully managed black box and won't manage data portability, Very small teams without an owner for evaluators, annotation queues, dashboards, and alerts.
Skip it if Skip Langfuse if you want a zero-ops logger you never tune — the Cloud tiers meter at $8/100k units past the included 100k, and self-hosting means running ClickHouse, Redis, and blob storage yourself.
GeneratedReal-world workflow fit
Concrete scenarios for the personas Langfuse actually fits — and what changes day-one when you adopt it.
A support agent starts returning wrong answers in production. You filter traces by userId and latency, open the failing trace, and walk the Timeline to find the retrieval step that returned nothing.
Outcome: You fix the retrieval filter and link the trace to the prompt version, so the change is traceable next time it regresses.
You attach an LLM-as-a-judge evaluator to a production rule, and Langfuse backfills scores onto recent historical observations while alerting you from the evaluator page when scores drop.
Outcome: Quality tracking starts from day one of the evaluator, not from day one of the project.
You open the usage chart on the Billing page to see units over time grouped by project, model a heavier month in the pricing calculator, and decide whether Pro's higher rate limits and retention justify $199/mo.
Outcome: You commit to a tier with a spend forecast instead of discovering overage at invoice time.
Use cases
- Debug an unexpected production agent behavior by replaying the exact trace, span by span, in the Langfuse UI.
- Compare two prompt versions on a dataset of real conversations and pick the winner side by side.
- Attach an LLM-as-judge evaluator to a rule and backfill scores onto recent historical observations.
- Track per-user LLM cost in a multi-tenant SaaS and bill accurately.
- Build golden datasets via human annotation queues for fine-tuning or evaluation.
- Create score and cost alerts from an evaluator page so quality regressions surface before customers notice.
- Ask Halo to run code over thousands of observations in the background sandbox to find what changed.
- Connect an IDE coding agent to traces and prompts through the MCP server, CLI, or SKILL.md.
Setup time & first value
How long it actually takes to get something useful out of Langfuse — broken out by persona, not the marketing-page minute.
With an existing OpenTelemetry setup or the Python/JS SDK, first traces usually land the same afternoon. Self-hosted Docker Compose is a quick local test; production Helm or Terraform on AWS/GCP/Azure with ClickHouse, Redis, and blob storage is a multi-day infrastructure task. Evaluators, annotation queues, and dashboards come after traces are flowing.
Switching to or from Langfuse
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From LangSmith: re-instrument via the Langfuse SDK or OpenTelemetry and replay stored runs into datasets for comparison.
- →From Helicone or a proxy logger: point LiteLLM or your gateway at Langfuse and keep proxy-based logging.
- →From a homegrown trace table: export to S3 blob storage or POST observations through the public API.
- →From Datadog LLM Observability: keep your OpenTelemetry instrumentation and redirect the exporter to Langfuse.
- ↗To LangSmith: export traces via the public API or S3 blob export and re-ingest through LangSmith's SDK.
- ↗To Datadog LLM Observability: reuse the OpenTelemetry instrumentation and send spans to Datadog's OTLP endpoint.
- ↗To self-hosted Langfuse: use the same APIs, since all platform features are MIT licensed and self-hostable.
Integrations
Resources & guides
- Documentationlangfuse.com
Overview
Langfuse is an open-source LLM engineering platform (GitHub) that helps teams collaboratively debug, analyze, and iterate on their LLM applications. All platform features are natively integrated to accelerate the development workflow.
- Guidelangfuse.com
Guides
End-to-end examples and resources to get started with Langfuse for LLM Tracing, Monitoring, Prompt Management, and more.
- Resourcelangfuse.com
Langfuse Academy
Understand why LLM engineering is different and how to navigate the full AI engineering lifecycle.
- Resourcelangfuse.com
Langfuse
Traces, evals, prompt management and metrics to debug and improve your LLM application.
- Resourcelangfuse.com
Langfuse
Traces, evals, prompt management and metrics to debug and improve your LLM application.
- Resourcelangfuse.com
Support
Overview of available support options for Langfuse.
- Resourcelangfuse.com
Overview
Helpful link from langfuse.com
Tutorials & Learning
YouTube returned 6 videos for “Langfuse”, and we withheld 5: 5 could not be judged, because “Langfuse” is a single word that other videos use for other things. Showing the 1 we can prove is about Langfuse.
Official links
What are the best Langfuse alternatives?
The closest alternatives to Langfuse we list are LangSmith, Helicone and MLflow. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| LangSmith | Freemium | Pricing not published | — |
| Helicone | Freemium | Pricing not published | — |
| MLflow | Free | Free | — |
Alternatives to Langfuse
View all
Tools that pair well with Langfuse
Common stack mates teams adopt alongside Langfuse, with the specific reason each pairing earns its keep.
LangSmith
LangSmith agent observability: trace, monitor, and evaluate every LLM run with SmithDB-backed trace search.
Helicone
Helicone is an AI gateway and LLM observability platform that routes, logs, and caches traffic across 100+ model providers.
MLflow
Open source LLM observability with OpenTelemetry tracing, evals, and prompt management for agents
Langfuse vs its closest rival
Langfuse's closest head-to-head is with Langgraph.
Our Langfuse vs Langgraph verdict written May 12, 2026 is more than 90 days old, so it is being re-reviewed against current data and is not shown.
Head-to-head comparisons
Langfuse vs Langgraph
Verdict being re-reviewed against current pricing and data.
Langchain vs Langfuse
If you need deep agent debugging with autonomous failure clustering and fix suggestions, LangSmith is the edge. If you want open-source flexibility, self-hosting, and unified prompt management plus observability, Langfuse is the pragmatic choice. Choose based on whether you need proactive root-cause analysis (LangChain) or full control and compliance via self-hosting (Langfuse).
GeneratedLangfuse vs Litellm
Verdict being re-reviewed against current pricing and data.
Langfuse vs Mlflow
Verdict being re-reviewed against current pricing and data.
Langfuse vs Promptfoo
Verdict being re-reviewed against current pricing and data.
Frequently asked questions
For makers of Langfuse
- Get a Verified Listing for Langfuse ($39): a written decision within 1 working day.
- Free listing for another tool: Free reviews are decided in order, within 3 weeks of the review starting.
Used Langfuse? Help shape our editorial sentiment research.