Both trace agents and score production traffic. Raindrop is a focused production monitor that clusters failures from plain-language signals; W&B Weave is a code-first eval and tracing toolkit, now part of CoreWeave Forge, that fits teams already training models on W&B.[1][6]
Raindrop is an independent, hosted monitor for agents in production that classifies live events with custom signals and groups recurring failures into issues; W&B Weave is a tracing, evaluation and monitoring toolkit from Weights & Biases, which CoreWeave acquired in May 2025. Since 30 September 2026 Weave is sold inside CoreWeave Forge, and hosted W&B accounts sign in through Forge. Weave suits teams already on W&B; Raindrop suits teams who want failures found for them.
| Decision factor | Raindrop | W&B Weave |
|---|---|---|
| Ownership | Independent; Series A led by CRV in September 2026 | CoreWeave, which completed its acquisition of W&B in May 2025; sold in CoreWeave Forge since 30 September 2026 |
| Centre of gravity | Production monitoring: Signals, Issue Detection, live Experiments | Evaluations, scorers and tracing in code, inside a model training platform |
| Offline evals | No hosted runner; local Workshop and early-access Simulations | Evaluation API with scorers, comparisons and leaderboards |
| Prompt tooling | None documented | Versioned prompts and a Playground |
| Run-time controls | None documented | Partial: Python guardrail scorers can block or modify inputs and outputs |
| Self-hosting | VPC deployment in beta for Enterprise and select partners | Kubernetes, with a W&B Platform install and a Weave-enabled licence |
| Free tier | Hobby: 1,000 events a month, 14-day retention | Forge Free: 1 GB of Weave ingestion a month |
| First paid tier | Pro: $299 a month plus $0.003 per event up to 1M events | Forge Pro: from $60 a month, 1.5 GB ingestion, teams under 50 people |
| Capability | Raindrop | W&B Weave | 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 | — | Partial | ✓ |
| Gateway or proxy in the request path | — | Not stated | — |
| Licence and deployment | |||
| Open source | Partial | Partial | Partial |
| Self-hosting | Partial | ✓ | Partial |
CoreWeave completed its acquisition of Weights & Biases on 5 May 2025. On 30 September 2026 it launched CoreWeave Forge, which bundles W&B Models, Weave and newer CoreWeave tools. Hosted W&B accounts now sign in through Forge, while self-hosted and dedicated-instance accounts are unchanged. Weave's pricing, docs and login now redirect to CoreWeave domains.
Two further changes affect anyone choosing Weave for agents. CoreWeave now recommends Agent Lens, a separate product in public preview that shares Weave's trace backend, for agent tracing and analysis. And Weave's docs describe its built-in monitors as a previous approach. Packaging is still moving, so confirm which product you are buying.
Raindrop is independent. It announced a Series A led by CRV on 17 September 2026, bringing total funding to $50M, and shipped a new classification system in August and Simulations in September.
Raindrop is organised around live traffic. Signals classify every new event against behaviours you describe in plain language, Issue Detection groups recurring failures, and Experiments compare signal rates between cohorts of real users. You do not write the judge: Raindrop's own models build the classifier, and custom judge prompts or models are not supported. A Triage Agent then investigates those issues in the web app, in Slack or over MCP, and the open-source Workshop turns failures into evals you run locally.
Weave is organised around code. You decorate functions with @weave.op to trace them, run the Evaluation API over a dataset with scorers, compare results to catch regressions and share leaderboards. Scorers can be LLM judges with prompt templates, functions, or Python classes, and annotation queues route traces to reviewers. Versioned prompts and a Playground sit alongside. For Weights & Biases users, all of this lives next to model training runs.
A team that wants to own every scorer will be more at home in Weave. A team that wants failures surfaced without building an eval suite first will be more at home in Raindrop.
As of 20 October 2026, Raindrop's Hobby plan is free for 1,000 events a month with one custom signal and 14-day retention. 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.
Weave is billed through CoreWeave Forge plans that combine seats, storage and Weave ingestion. Forge Free includes 1 GB of Weave ingestion a month and up to five model seats. Forge Pro starts at $60 a month with 1.5 GB of ingestion and ten seats, is limited to teams under 50 people, and lists overage at $0.10 per MB. Weave's allowance is measured in data volume rather than a count of traces, so estimate it from the size of your traces. Check Raindrop's plans and CoreWeave Forge pricing.
Weave has guardrails: scorers called from Python application code can block or modify inputs and outputs before they reach users. The TypeScript SDK does not support them, and they check content rather than approving or refusing an agent's tool actions. Raindrop documents no blocking or approvals; it analyses events after ingestion and alerts, and its PII Guard redacts data on arrival rather than controlling what the agent does.
On instrumentation, both accept OpenTelemetry. Raindrop ingests OTLP traces over HTTP as JSON or protobuf and works with the OpenTelemetry Collector, but takes no gRPC, metrics or logs. Weave accepts OTLP over HTTP as protobuf, reads the OTel GenAI, OpenInference and OpenLLMetry conventions, and has turned its OTel integrations on by default since June 2026. Both track model cost, though Weave's cost tracking works from Python only. Self-hosting is narrow on both sides: Raindrop's VPC deployment is in beta, and Weave needs a W&B Platform installation and a Weave-enabled licence.
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 and a kill switch for a run, an agent or a team. Weave's Python guardrails check content in your code, and Raindrop has no run-time controls; neither holds a tool action for approval. Prefactor has no prompt playground or prompt registry, is not tied to a model training platform, and its platform is hosted. 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.