Langfuse vs. LiteLLM Lens

This guide outlines the key differences between Langfuse and LiteLLM Lens to help engineering teams choose the right AI engineering platform. All Lens facts were checked against the public LiteLLM docs and the Lens GitHub repository in October 2026.

TL;DR

• A shipped product with an open-source core, on Cloud or self-hosted. Langfuse's core tracing, evaluation, prompt management, and APIs are MIT-licensed with versioned releases on Cloud and self-hosted; additional Enterprise features require a license, and Enterprise SSO is in the free distribution. Lens launched on October 1, 2026 as early access; it is MIT-licensed, self-run only, and its releases page says to build from source until a release is published.
• 3 years of history, with the platform on top of ClickHouse. Both products store traces in ClickHouse. Langfuse is part of ClickHouse, keeps 3 years on Pro, and puts APIs, dashboards, SQL when self-hosting, and scheduled exports on Pro with the Teams add-on, Enterprise, and self-hosted on that table. Lens defaults to 14 days of retention, configurable by the operator, on a single ClickHouse server you operate.
• Published pricing. Langfuse Cloud publishes Core at $29, Pro at $199 with unlimited users, and Enterprise at $2,499; self-hosting is free. Lens has no license fee and no hosted plan; the cost is the service, its ClickHouse, and the analysis-model spend each investigation caps.
• The loop runs in one project, prompts included. Langfuse judges and code evaluators score incoming observations, prompt experiments run in the UI and applications through the SDK, annotation queues are on every plan, and prompt labels carry model and tool config. Lens investigates a sample of stored traces with an analysis model and gates pull requests through a GitHub Action; prompt management and annotation queues are not documented.
• Keep LiteLLM, switch the quality record the same day. Add the Langfuse SDK or OpenTelemetry exporter next to the proxy integration; exported Lens datasets import via SDK, API, or CSV. Most teams see first traces in Langfuse the same day.

Choose Langfuse if the system of record has to include review, experiments, and prompt versions: an open-source core on Cloud or self-hosted including air-gapped, a published price with 3 years of history on Pro, and observations in ClickHouse you can query and export, with the LiteLLM gateway kept for routing, keys, and budgets. Proven at 21 of the Fortune 50.

Choose Lens if you are already running LiteLLM, want its investigator over agent traces on infrastructure you operate, and can build and pin it from source. You accept early-access status with no published release, a 14-day default retention on one ClickHouse server, no hosted option, and a loop without prompt management or annotation queues.

---

Open source and deployment

Langfuse's core tracing, evaluation, prompt management, and APIs are MIT-licensed and available on Cloud or self-hosted; additional Enterprise features require a license. Self-hosting runs the UI, login, and data in your environment on every tier, including air-gapped operation, and Enterprise SSO and organization-level RBAC ship in the free distribution. Langfuse Cloud is covered by SOC 2 Type II and ISO 27001 audits, is GDPR compliant, and signs a BAA from the Pro plan in the HIPAA region. Langfuse is used by 21 of the Fortune 50 and 129 of the Fortune 500.

Lens launched on October 1, 2026 and its overview is labeled early access. The repository is MIT-licensed outside an enterprise/ directory. Lens runs as its own service with ClickHouse, standalone behind an admin token or inside the LiteLLM sidebar using LiteLLM's identity, and its releases page states that, as of October 9, 2026, there is no published GitHub release and the quickstart builds from source.

---

Storage and retention

ClickHouse has been open source since June 15, 2016 (Apache 2.0) and is today the most popular open-source analytical database, with 2,000+ contributors. Langfuse Cloud and self-hosted run the same ClickHouse architecture, so you can move between them without locking into a closed store.

Both products store traces in ClickHouse. Langfuse works directly with ClickHouse's core team as part of ClickHouse and runs it as the primary store for the whole application on Cloud and self-hosted, on a managed cluster, ClickHouse Cloud, or your own. Every LLM call, tool execution, and agent step is written once to a wide, immutable observations table (details). High-volume reads go through the Observations and Metrics APIs, scheduled exports (Pro with the Teams add-on, Enterprise, and self-hosted) land JSONL or Parquet in your own bucket, dashboards and Pulse read the same table, and self-hosters query ClickHouse in SQL.

Lens stores traces and its own state in a ClickHouse database you operate. Its configuration reference sets trace retention with AGENT_TRACING_RETENTION_DAYS, default 14 days, and its storage page supports one ClickHouse server shared by the Lens replicas, with multi-server load balancing and automatic failover listed as unsupported.

---

Pricing

Langfuse Cloud publishes its prices: Core at $29, Pro at $199 with 3 years of history and unlimited users, Enterprise at $2,499, plus graduated unit pricing where a unit is one trace, observation, or score. Judge inference runs on your own model connection, and scheduled bucket exports and Cloud SSO are the Teams add-on. Self-hosted Langfuse OSS has no license fee; you run the infrastructure.

Lens has no license fee and no hosted plan. You run the service and its ClickHouse, and every investigation carries its own monthly spending limit in USD for the analysis model, with provider charges on top; routing that analysis through a LiteLLM key applies that key's budgets as well.

See Langfuse pricing or talk to us about Enterprise volumes.

---

The improvement loop

Langfuse runs the loop in one project. LLM-as-a-judge and code evaluators score incoming observations and backfill history, alerts fire on those metrics, and Halo investigates traces and creates datasets, prompts, and dashboards on request; it is on every Cloud plan and in public beta for self-hosted. A failure becomes a dataset item, a prompt experiment in the UI or a full-application experiment through the SDK that you compare to a baseline, a row in an annotation queue, and a prompt label that carries model and tool config your application reads at runtime, with a CI gate on the result.

Lens investigations take a time window, a sample percentage, and a description of what the agent should do, send the selected traces to an analysis model, and group findings with quotes linked to the original step, on demand or on a schedule. Signals flag traces that match yes/no questions through a separate evaluation provider, saved runs become datasets, and the repository documents PR evals that run the agent in a GitHub Action and fail the check on a regression. Prompt management and annotation queues are not documented.

---

Using Langfuse and LiteLLM together

LiteLLM stays the gateway for routing, virtual keys, and budgets. The LiteLLM integration records proxy traffic in Langfuse, and the Langfuse SDK or OpenTelemetry adds the retrieval, tool, and application spans the proxy does not see. A Langfuse Gateway is on the public roadmap; until it ships, routing stays with LiteLLM or another gateway.

If you trial Lens beside Langfuse, decide which store holds datasets and human labels before both accumulate a partial copy; running two investigators over the same traffic doubles analysis-model spend.

---

Which should you choose

Choose Langfuse if you want a shipped, independent platform for the LLM application: an open-source core on Cloud or self-hosted including air-gapped with SSO and organization RBAC in the free distribution, prompt experiments in the UI and application experiments through the SDK, annotation queues, prompt labels, and CI gates in one project, 3 years of observations in ClickHouse you can query and export, and a published price.

Choose Lens if you are already running LiteLLM, want its investigator over agent traces on infrastructure you operate, and can build and pin it from source.

Keep LiteLLM for the gateway either way. See Using Langfuse and LiteLLM together.

---

Why teams move away from Lens

The same reasons recur when teams evaluate both.

• The artifact you deploy matters. Lens's releases page says to build from source until a release is published, and a gateway upgrade does not carry a Lens service deployed beside it. Langfuse Cloud and the self-hosted release are the shipped product, with the same open-source core on both.
• Retention default and topology. Lens defaults to 14-day retention, configurable by the operator. Its documented topology supports one ClickHouse server shared by Lens replicas; multi-server load balancing and automatic failover are unsupported. Langfuse Pro keeps 3 years on a managed or clustered ClickHouse, with APIs, dashboards, and scheduled exports on Pro with the Teams add-on, Enterprise, and self-hosted on the same table.
• Prompts and review are missing from the loop. Lens investigates, flags, and gates pull requests; it documents no prompt management or annotation queue. Langfuse runs judges, experiments, queues, prompt labels, and CI in one project.
• No hosted option and no published price. Lens is self-run only, with analysis-model spend capped per investigation on a ClickHouse you operate. Langfuse Cloud publishes Core, Pro, and Enterprise with unlimited users, and self-hosting is free.

---

Why teams choose Langfuse

A pattern repeats across production adopters: start where governance requires, often self-hosted, then grow into the platform.

• Canva runs a multi-agent support system for a product with 250 million monthly active users. LLM-as-a-judge evaluators score the systems across 15–20 metrics, and domain experts build evaluators without an engineering handoff. They self-hosted first, then moved to Cloud.
• SumUp rolled AI support to 35+ markets and deflected almost 50% of conversations, cutting external BPO cost by 30%. They started self-hosted for the proof of concept, then moved to Langfuse Cloud.
• Merck runs about 80 of 300+ GenAI use cases on Langfuse, with 200+ people building on it. They self-host for data sovereignty and provision projects via API.

---

Switching from Lens

Leave the gateway in place and move the quality record. Most teams see first traces in Langfuse the same day.

• Instrumentation: add the Langfuse SDK or point the OpenTelemetry exporter at the OTLP endpoint, plus the LiteLLM integration for proxy spans.
• Datasets: import exported Lens datasets through the Langfuse SDK, API, or CSV uploader; expected responses become ground truth. Create judges in evaluators and prompts in prompt management.
• History: stays in the Lens ClickHouse database for its retention window. New traces accumulate in Langfuse from the cutover.

Talk to us if you want help planning the switch.

---

Start free: Cloud or self-host

Langfuse Cloud Hobby includes 50k units/month, no credit card. Or self-host the MIT stack. The example project is populated before you instrument anything.

---

FAQ

---

---