Both trace agents, run experiments and score production traffic with LLM judges and code. Langfuse is the open-source option you can run yourself for free, with prompt versioning built in; Fiddler is a commercial platform with real-time guardrails and classic ML monitoring, sold self-serve per trace or through Enterprise. Licence and run-time needs usually decide it.[1][4]
Langfuse is MIT-licensed and free to self-host, with tracing, online evals and prompt versioning; ClickHouse acquired it in January 2026 and Langfuse says its licence is unchanged. Fiddler is a proprietary platform that adds real-time guardrails and classic ML monitoring, with self-hosting on Enterprise only and no prompt tooling. Langfuse does not act during a run; Fiddler scores content in real time for your application or a gateway to block.
| Decision factor | Fiddler | Langfuse |
|---|---|---|
| Ownership | Independent; $30M Series C in January 2026 | Acquired by ClickHouse in January 2026; licence and self-hosting unchanged, per Langfuse |
| Licence | Proprietary platform; Python client is Apache-2.0 | MIT, except add-on features in the ee folders |
| Self-hosting | Enterprise: VPC, on-premises or SageMaker | Free, via Docker or Kubernetes |
| Prompt tooling | None | Prompt versioning with label-based deployment and a playground |
| Run-time checks | Partial: real-time content scores; inline blocking through a third-party AI gateway | None; evaluation happens after the fact |
| Classic ML monitoring | Yes: drift, performance, explainability, fairness | Not documented |
| Free tier | Guardrails-only free plan, by demo request | Hobby: 50k units a month, 30 days of data; self-hosting free |
| First paid tier | Developer: $0.002 per trace, hosted only | Core: $29 a month, 100k units |
| Capability | Fiddler | Langfuse | Prefactor |
|---|---|---|---|
| Observe | |||
| LLM call tracing | ✓ | ✓ | ✓ |
| Multi-step agent tracing (tool calls, nested spans) | ✓ | ✓ | ✓ |
| Cost tracking | Partial | ✓ | ✓ |
| OpenTelemetry support | ✓ | ✓ | ✓ |
| Evaluate | |||
| Offline evals (datasets, experiments) | ✓ | ✓ | ✓ |
| Online evals on production traffic | ✓ | ✓ | ✓ |
| LLM-as-a-judge scoring | ✓ | ✓ | ✓ |
| Code-based evaluators | ✓ | ✓ | ✓ |
| Human review and annotation | Partial | ✓ | ✓ |
| Prompt versioning and playground | — | ✓ | — |
| Act on findings | |||
| Runtime controls on live agent actions | Partial | — | ✓ |
| Gateway or proxy in the request path | — | — | — |
| Licence and deployment | |||
| Open source | Partial | ✓ | Partial |
| Self-hosting | ✓ | ✓ | Partial |
Langfuse publishes its platform under the MIT licence, apart from a few add-on features in licence-keyed ee folders, and documents a free self-hosted path on Docker or Kubernetes backed by Postgres, ClickHouse, Redis and object storage. ClickHouse acquired Langfuse in January 2026; Langfuse said its roadmap and licence would not change and that self-hosting stays a first-class path. Its free self-hosted version includes online evaluator rules on production traffic, not only tracing.
Fiddler's platform is proprietary; its Python client and Claude Code plugin are Apache-2.0. Self-hosting is part of the Enterprise plan, as VPC, on-premises or SageMaker deployment in your AWS account, priced through sales. If your traces cannot leave your network and you have no budget for a contract, Langfuse is the one that works. If you want a vendor to support an on-premises install under contract, Fiddler sells that as part of Enterprise.
Langfuse links prompt versions to traces, deploys them by label without code changes, compares metrics across versions and has a playground. Fiddler has no prompt versioning; its Prompt Specs are templates for judge evaluators. A team that changes prompts often will want Langfuse or a separate prompt tool alongside Fiddler.
Fiddler, in turn, can act on content in real time. Fiddler Guardrails score requests and responses for harmful content, faithfulness, personal data and secrets, and your application decides whether to block; inline blocking or redaction works when Fiddler sits behind Kong AI Gateway, AgentGateway, LiteLLM or Portkey. Langfuse does not block anything. Its documentation says run-time protection comes from separate libraries and that Langfuse evaluates those measures after the fact. Neither documents approval steps or policy checks on an agent's tool calls.
Fiddler monitors classic ML models for drift, performance, explainability and fairness in the same platform as its LLM and agent traces, which suits an organisation where one team owns both. Langfuse focuses on LLM and agent applications.
Langfuse ships SDKs for Python and TypeScript and accepts OTLP over HTTP, though not gRPC. Fiddler's native agent SDKs are Python only, covering LangGraph, LangChain, Strands Agents and Google ADK, and it also accepts OTLP. One date for existing Langfuse Cloud users: the service becomes v4-only on 16 November 2026, which may need migration work.
The evaluation tools are closer than the licences suggest. Langfuse added code evaluators in Python or TypeScript in May 2026, a GitHub Action that fails a CI run below a score threshold, and in September 2026 evaluator backfills over up to six months of past observations; reviewers work through annotation queues. Fiddler's Evals SDK ships 13 pre-built evaluators and wraps any Python function as one, its evaluator rules also backfill, and since September 2026 it hosts its own Ministral 3 8B judge model. Fiddler's human review is lighter: span annotations entered public preview in August 2026, with no review queue documented.
As of 20 October 2026, Langfuse Hobby is free for 50k units a month with 30 days of data, Core is $29 a month for 100k units then $8 per 100k, Pro is $199 a month and Enterprise $2,499 a month or custom; self-hosting is free apart from your infrastructure. A Langfuse unit is a trace, observation or score, so one agent run can use several. Fiddler's Developer plan charges $0.002 per trace, about $200 for 100,000 traces a month; its free plan covers guardrails only and is reached through a demo request.
Langfuse Pro includes three years of data access, with a Teams add-on at $300 a month. Fiddler's Enterprise plan adds VPC and on-premises deployment, enterprise guardrails and a named customer success contact.
Check Fiddler's pricing and Langfuse's pricing against 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 on tool calls and data access, approval holds for risky actions, and a kill switch for a run, an agent or a team. Langfuse documents no run-time controls, and Fiddler's guardrails check content rather than tool actions. Prefactor has no prompt playground or prompt registry and its platform is hosted, so if free self-hosting or prompt versioning is the requirement, Langfuse 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.