Both trace agents, run experiments, score production traffic, queue traces for review and version prompts. Langfuse is the platform you can run yourself for free; Confident AI is the deeper evaluation toolkit, with an open-source metric library and a hosted platform. Self-hosting needs usually decide it; if not, eval depth does.[1][4]
Langfuse open-sources the whole platform (MIT, apart from licensed add-ons) and lets you self-host it for free; Confident AI open-sources its evaluation framework, DeepEval, and keeps the platform proprietary, with self-hosting on Enterprise only. ClickHouse acquired Langfuse in January 2026 and Langfuse says its licence and self-hosting are unchanged. Confident AI goes further on evaluation, with 40+ metrics, red teaming and simulation; neither blocks a live agent action.
| Decision factor | Confident AI | Langfuse |
|---|---|---|
| Ownership | Independent; Y Combinator Winter 2025 company | Acquired by ClickHouse in January 2026; licence and self-hosting unchanged, per Langfuse |
| What is open source | DeepEval and confident-trace (Apache-2.0); platform proprietary | The platform (MIT), except add-ons in the ee folders |
| Self-hosting | Enterprise only, licence key and private images | Free, via Docker or Kubernetes |
| Built-in metrics | 40+ DeepEval metrics, including agent, multi-turn, RAG and MCP | LLM-as-a-judge and code evaluators you define |
| Online evals | Yes; referenceless metrics only during tracing | Yes, with sampling and backfills up to six months |
| Red teaming and simulation | Yes | Not documented |
| Free tier | 2 seats, 1 GB-month of trace data, 5 test runs a week | Hobby: 50k units a month, 30 days of data |
| First paid tier | Starter: $200 a month per organisation | Core: $29 a month, 100k units |
| Capability | Confident AI | Langfuse | 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 | — | — | ✓ |
| Gateway or proxy in the request path | — | — | — |
| Licence and deployment | |||
| Open source | Partial | ✓ | Partial |
| Self-hosting | Partial | ✓ | Partial |
Both vendors publish open-source code, but not the same part. Langfuse publishes its platform under the MIT licence, apart from a few add-on features in licence-keyed ee folders, and documents free self-hosting on Docker Compose or Kubernetes, backed by Postgres, ClickHouse, Redis and object storage. ClickHouse acquired Langfuse in January 2026; Langfuse said the roadmap and licence would not change, and it has kept shipping, with code evaluators in May and evaluator backfills in September.
Confident AI publishes DeepEval, its evaluation framework, and confident-trace, an OpenTelemetry tracing SDK, both under Apache-2.0. The platform that stores traces, runs online evals and hosts the UI is proprietary. Self-hosting it needs an Enterprise licence key and private container images. If traces must stay on your servers and there is no budget for an Enterprise contract, Langfuse is the one that works.
Langfuse Cloud runs in EU, US, Japan and HIPAA regions, and self-hosting lets you choose any location. Confident AI hosts in the US or EU on every plan, with custom data residency on request; it shows SOC 2 Type II, provides the report and a HIPAA BAA on the Team plan and above, and lists SSO, RBAC and audit logs. If you need a hosted region in Japan, Langfuse lists one and Confident AI does not.
Langfuse gives you the machinery to evaluate: evaluator rules on live traffic with sampling, LLM-as-a-judge and Python or TypeScript code evaluators, experiments on datasets or production traces, a GitHub Action that fails CI below a threshold, and annotation queues. You write most of the judging logic yourself.
Confident AI ships the judging logic too. DeepEval has more than 40 ready metrics covering agents, multi-turn conversations, RAG, MCP and multimodal output, with G-Eval and DAG custom metrics, and the platform adds red teaming, multi-turn and voice simulation and tool failure analytics. One limit to note: its online evals run only referenceless metrics during tracing, so metrics that need an expected answer run offline.
Both version prompts. Langfuse deploys prompt versions by label without code changes, links prompts to the traces that used them and compares metrics across versions. Confident AI's Prompt Studio has commits, versions and environment labels, pulls prompts by label in code and runs side-by-side experiments across prompt versions and models.
These two combine more easily than most pairs. DeepEval runs locally without the Confident AI platform, and Langfuse accepts custom scores through its API and SDKs, so a team can trace and self-host on Langfuse while running DeepEval metrics in its own code and pushing the results to Langfuse as scores. You lose Confident AI's hosted reports and red teaming, but keep a fully open-source stack.
Existing Langfuse Cloud users should also note that Langfuse Cloud becomes v4-only on 16 November 2026, which may need migration work whichever way you go.
The reverse also works in part: Confident AI's OTLP endpoint reads OpenLLMetry attributes, which Langfuse also accepts, so instrumentation built on OpenLLMetry can be pointed at either backend while you compare them.
As of 20 October 2026, Langfuse prices per unit ingested: Hobby is free for 50k units a month, Core is $29 a month for 100k units then $8 per 100k, Pro is $199 a month and Enterprise is $2,499 a month or custom. Self-hosting is free apart from your infrastructure. Confident AI prices per organisation with trace storage in GB-months: Free covers 2 seats and 1 GB-month, Starter is $200 a month with unlimited seats, Team is $2,000 a month, and extra storage is $1 per GB-month.
Langfuse's entry price is much lower; Confident AI's Starter includes unlimited seats. Compare Confident AI's pricing and Langfuse's pricing on your own volumes.
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 and acts on what it finds through runtime policies, approval holds and a kill switch scoped to a run, an agent or a team. Both Confident AI and Langfuse evaluate after the fact: Langfuse says run-time protection comes from separate libraries, and Confident AI's threshold checks run on recorded data. Prefactor has no prompt playground or prompt registry and its platform is hosted, so if free self-hosting or an open-source metric library is the requirement, Langfuse or DeepEval is the better fit. 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.