Both trace agents and score production traffic. Langfuse is the open-source option you can run yourself, with evaluators you define and prompt versioning; Raindrop is a hosted monitor that drafts classifiers from plain-language descriptions and groups recurring failures into issues. If you need to self-host, Langfuse is the only one of the two that does it without a sales process.[1][4]
Langfuse is an open-source (MIT) platform for tracing, evaluation and prompt versioning that you can self-host for free, while Raindrop is a proprietary hosted monitor that classifies live agent traffic and groups recurring failures into issues. ClickHouse acquired Langfuse in January 2026, and Langfuse says its licence and self-hosting are unchanged; Raindrop is independent. Langfuse lets you define your own judges and code evaluators; Raindrop builds its classifiers with its own models.
| Decision factor | Langfuse | Raindrop |
|---|---|---|
| Ownership | Acquired by ClickHouse in January 2026; licence and self-hosting unchanged, per Langfuse | Independent; Series A led by CRV in September 2026, $50M raised in total |
| Licence | MIT, except add-on features in the ee folders | Proprietary platform; Workshop and several SDKs are MIT |
| Self-hosting | Free, via Docker or Kubernetes | Beta: Enterprise VPC deployment for select partners |
| Evaluators | LLM judges and Python or TypeScript code evaluators you write | Signals drafted from a description; no custom judge prompt or model |
| Production scoring | Evaluator rules with sampling, plus backfills over up to six months | Signals on every new event; Issue Detection groups recurring failures |
| Prompt tooling | Versioning, label-based deployment and a playground | None documented |
| Free tier | Hobby: 50k units a month, 30 days of data | Hobby: 1,000 events a month, 14-day retention |
| First paid tier | Core: $29 a month for 100k units | Pro: $299 a month plus $0.003 per event |
| Capability | Langfuse | Raindrop | Prefactor |
|---|---|---|---|
| Observe | |||
| LLM call tracing | ✓ | ✓ | ✓ |
| Multi-step agent tracing (tool calls, nested spans) | ✓ | ✓ | ✓ |
| Cost tracking | ✓ | ✓ | ✓ |
| OpenTelemetry support | ✓ | ✓ | ✓ |
| Evaluate | |||
| Offline evals (datasets, experiments) | ✓ | Partial | ✓ |
| Online evals on production traffic | ✓ | ✓ | ✓ |
| LLM-as-a-judge scoring | ✓ | Partial | ✓ |
| Code-based evaluators | ✓ | Partial | ✓ |
| Human review and annotation | ✓ | Partial | ✓ |
| 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 |
Langfuse publishes its platform under the MIT licence, apart from add-on features in licence-keyed ee folders, and documents free self-hosting on Docker Compose for testing or Kubernetes for production, 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 option, and it has shipped steadily since, with code evaluators in May and evaluator backfills in September.
Raindrop is hosted and proprietary. Its self-hosting is a beta Enterprise VPC deployment rolling out to select partners. The parts you can run yourself are Workshop, an MIT-licensed local debugger, and several SDKs. If traces must stay inside your network today, Langfuse is the practical option of the two.
One date for current Langfuse Cloud users: Cloud becomes v4-only on 16 November 2026, which may need migration work.
Langfuse expects you to define quality. LLM-as-a-judge evaluators and, since 28 May 2026, Python or TypeScript code evaluators run on live observations or on experiment runs and are stored as scores. Annotation queues let reviewers score traces against defined configs, and experiments run against datasets, production traces or local data, with a GitHub Action that fails CI below a threshold.
Raindrop takes the definition off your hands. You describe a behaviour in plain language and it drafts a classifier, a Signal, that runs on every new event and backfills a sampled 90 days; you tune it by labelling samples as Match or Not Match. Issue Detection groups recurring failures, a Triage Agent investigates them in the web app, Slack or over MCP, and Experiments compare signal rates between cohorts of live traffic.
The trade-off is control against setup time. Langfuse shows you the exact judge prompt or code behind every score. Raindrop gets you classifiers faster but you cannot bring your own judge prompt or model, and it has no hosted offline eval runner: pre-release work goes through Workshop locally or Simulations, which is in early access.
Both accept OpenTelemetry over HTTP and neither accepts gRPC. Langfuse acts as an OTLP backend, names OpenLLMetry and OpenLIT as compatible and maps OpenInference-style attributes. Raindrop accepts traces with gen_ai attributes, but not metrics or logs, and works with the OpenTelemetry Collector. Neither sits in the request path as a gateway.
Prompt versioning is a clear gap on Raindrop's side. Langfuse versions prompts, deploys them by label without code changes, links them to traces and compares metrics across versions. Raindrop can compare prompt versions only through Experiments on production traffic, or replay a trace locally in Workshop with an adjusted system prompt.
For multi-step agents, Langfuse nests observations, groups multi-turn work into sessions and has drawn an agent graph view since July 2026. Raindrop follows every message, tool call and decision in a run, with integrations for many agent frameworks, most labelled beta. On cost, Langfuse records token usage and cost per generation, ingested or inferred from model definitions, with per-user cost through a user ID. Raindrop calculates spend by model, provider and project from a versioned catalogue of public model rates.
As of 20 October 2026, Langfuse charges per unit ingested, counting traces, observations and scores. 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, and Pro is $199 a month. Self-hosting removes the subscription but adds the infrastructure you run. Raindrop's Hobby plan is free for 1,000 events a month, after which ingestion stops until the next month. Pro is $299 a month plus $0.003 per event up to 1M events, then $0.002, with unlimited retention and a 14-day trial.
The units differ, so model your own traffic against Langfuse's pricing and Raindrop's plans. On published entry prices, Langfuse's first paid tier is lower.
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 acts on a failing score through runtime policies, approval holds and a kill switch scoped to a run, an agent or a team. Neither Langfuse nor Raindrop documents run-time controls: Langfuse says protection comes from separate libraries, and Raindrop alerts after ingestion. 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.