These are the two closest products in this set: open source, self-hostable, with tracing, experiments, online evals, annotation queues and prompt versioning. Opik adds guardrails and automated prompt optimisation; Langfuse offers a larger free cloud tier, longer paid retention and evaluator backfills over past data. Specific features, not categories, decide this one.[1][4]
Langfuse and Opik differ less than most pairs: both are open source and free to self-host with online evaluation included, Langfuse under MIT (apart from licence-keyed add-ons) and Opik, from Comet, under Apache-2.0. ClickHouse acquired Langfuse in January 2026, and Langfuse says its licence and roadmap are unchanged. Opik adds guardrails your code can call and a prompt optimiser; Langfuse has no run-time checks but a larger free cloud tier and evaluator backfills.
| Decision factor | Langfuse | Opik |
|---|---|---|
| Ownership | Acquired by ClickHouse, January 2026; no planned licence changes | Comet ML, independent; reports $70M in total funding |
| Licence | MIT, except add-ons in the ee folders | Apache-2.0 for the core platform and SDKs |
| Self-hosted stack | Postgres, ClickHouse, Redis or Valkey, S3-compatible storage | Kubernetes for production; local Docker not production-ready |
| Run-time checks | None; evaluation happens after the fact | Partial: guardrails your code calls; they fail closed |
| Prompt tooling | Versioning, labels, playground | Versioning, environments, Agent Playground, Prompt Optimizer |
| OpenTelemetry | OTLP over HTTP; gRPC not supported | OTLP over HTTP; gRPC not supported |
| Free cloud tier | Hobby: 50k units a month, 30 days of data | Free Cloud: 25k spans a month, 60-day retention |
| First paid tier | Core: $29 a month, 100k units, 90 days of data | Pro Cloud: $19 a month, 100k spans, 60-day retention |
| Capability | Langfuse | Opik | Prefactor |
|---|---|---|---|
| Observe | |||
| LLM call tracing | ✓ | ✓ | ✓ |
| Multi-step agent tracing (tool calls, nested spans) | ✓ | ✓ | ✓ |
| Cost tracking | ✓ | ✓ | ✓ |
| OpenTelemetry support | ✓ | ✓ | ✓ |
| Evaluate | |||
| Offline evals (datasets, experiments) | ✓ | ✓ | ✓ |
| Online evals on production traffic | ✓ | ✓ | ✓ |
| LLM-as-a-judge scoring | ✓ | ✓ | ✓ |
| Code-based evaluators | ✓ | ✓ | ✓ |
| Human review and annotation | ✓ | ✓ | ✓ |
| Prompt versioning and playground | ✓ | ✓ | — |
| Act on findings | |||
| Runtime controls on live agent actions | — | Partial | ✓ |
| Gateway or proxy in the request path | — | Partial | — |
| Licence and deployment | |||
| Open source | ✓ | ✓ | Partial |
| Self-hosting | ✓ | ✓ | Partial |
Both platforms publish their code and document a free self-hosted path, so the choice comes down to details. Langfuse is MIT-licensed apart from its ee folders, which hold add-on features that need a licence key, such as UI customisation. Self-hosting runs on Docker Compose for testing and on Kubernetes, AWS, Azure or GCP for production, backed by Postgres, ClickHouse, Redis or Valkey and S3-compatible storage.
Opik is Apache-2.0 throughout, and its documentation says the self-hosted version includes every feature except user administration. The local install script uses Docker and is described as not production-ready; production deployments use Kubernetes. If you want Opik's guardrails on your own servers, they need a separate guardrails server, with a GPU recommended. One default to know: since September 2026 the Opik SDK sends anonymous feature-usage analytics, which you can turn off with OPIK_ANALYTICS_ENABLE=false.
ClickHouse acquired Langfuse in January 2026. Langfuse said its roadmap would not change, its licence would not change and self-hosting would stay a first-class option, and it has kept shipping: rebuilt experiments in April, code evaluators in May, a 1.0 command-line tool in August, evaluator backfills in September and release v4.64.0 on 19 October. Existing cloud users should note that Langfuse Cloud becomes v4-only on 16 November 2026, which may need migration work.
Comet, which builds Opik, is independent. Opik 2.0 shipped in June 2026 with an environment-tagged Prompt Library and resumable evaluations; July added a per-evaluation spend cap for judges, September an import command for traces of any age, and October a native Ollama integration and annotation queues that fill from feedback-score conditions.
Run-time checks are the clearest split. Langfuse does not block or enforce anything on live requests; its documentation says protection comes from separate libraries and Langfuse evaluates those measures afterwards, and its threshold alerts notify rather than block. Opik ships guardrails for PII, topics, prompt injection, judge checks and custom classifiers as a blocking validate() call that fails closed. Your code must call them, PII and topic checks are English only, and there is no approval workflow.
On evaluation, Langfuse can backfill evaluator scores over up to six months of past observations, offers a low-cost classifier judge, and runs experiments in CI through a GitHub Action that fails below a score threshold. Opik offers more than 30 built-in metrics, an Agent as a Judge that reads traces with tools, Test Suites written as plain-text assertions and Ollie, an assistant that analyses traces and proposes fixes. Both version prompts with labels or environments and a playground; Opik adds a Prompt Optimizer with six algorithms.
Instrumentation is close to interchangeable. Both accept OTLP over HTTP and neither supports gRPC. Langfuse maps OpenInference-style attributes; Opik does not state which conventions it reads.
The cloud plans count different things. As of 20 October 2026, Langfuse charges per unit, where traces, observations and scores all count: Hobby is free for 50k units a month with 30 days of data, Core is $29 a month for 100k units and 90 days of data, then $8 per 100k, and Pro is $199 a month with three years of data. Opik charges per span: Free Cloud covers 25k spans a month with 60-day retention, and Pro Cloud is $19 a month for 100k spans, then $5 per 100k, with retention up to 400 days at $29 per 100k spans.
Self-hosting either one removes the subscription but adds infrastructure you run, which for Langfuse includes a ClickHouse database. Because Langfuse counts scores as units, heavy online evaluation adds to its bill in a way that span counts do not show. Model both on your own traffic using Langfuse's pricing and Opik's pricing.
We sell the layer this section describes. Read it with that in mind.
Stated plainly, since Prefactor publishes this page: Prefactor is an agent evaluation platform that scores each step of a run, not only the final answer, and can act on what it finds through runtime policies, approval holds and a kill switch scoped to a run, an agent or a team. Langfuse has no run-time checks, and Opik's guardrails block only where your code calls them, without approval holds or tool-call policies. Prefactor has no prompt playground or prompt registry and its platform is hosted, so if free self-hosting or prompt iteration is the requirement, Langfuse or Opik is the better pick. See the full comparison of 15 platforms.
Reviewed against public product and documentation pages on October 20, 2026. If a vendor has changed a feature, product name, or positioning since then, send a correction and we will update it. Numbered source links in the page body point to the ordered sources below.
Prefactor scores each step of every agent run and can hold or stop an action when a score says it should not happen. Start on the free tier: 5,000 spans a month, no credit card required.
Prefactor helps teams observe, evaluate, and improve their AI agents in production, across every framework and provider.