Both are Python-first, accept OpenTelemetry and cover tracing, evals and prompt versioning. Logfire puts agent traces next to your application's HTTP and database spans, with a gateway in front of model calls; W&B Weave, now sold through CoreWeave Forge, puts them next to your model training runs.[1][6]
Pydantic Logfire is OpenTelemetry observability for your whole application, with evals, prompt versioning and an AI gateway; W&B Weave is a tracing and evaluation toolkit for LLM apps that sits beside Weights & Biases' model training tools. CoreWeave acquired W&B in May 2025, and since 30 September 2026 Weave is sold inside CoreWeave Forge. Both are Python-first; what sits next to your traces is the real difference.
| Decision factor | Logfire | W&B Weave |
|---|---|---|
| Ownership | Pydantic, independent; Series A led by Sequoia in 2024 | CoreWeave, which completed its acquisition of W&B in May 2025; sold in CoreWeave Forge since 30 September 2026 |
| What sits beside the traces | The rest of your application: HTTP, database, logs, metrics | W&B model training, experiment tracking and registry |
| Evals | pydantic-evals datasets, experiments and live evaluations | Evaluation API, scorers, leaderboards and monitors |
| Human review | Run annotations in beta; queues in early access | Annotation queues with boolean, numeric, categorical or text fields |
| Run-time checks | Partial: gateway guardrails and spend caps on routed model calls | Partial: Python guardrail scorers in your code |
| Self-hosting | Enterprise only, Kubernetes via Helm | Kubernetes, with a W&B Platform install and a Weave-enabled licence |
| Free tier | Personal: 10M records a month, 1 seat | Forge Free: 1 GB Weave ingestion a month |
| First paid tier | Team: $49 a month, 10M records, 5 seats | Forge Pro: from $60 a month, 1.5 GB ingestion |
| Capability | Logfire | 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) | ✓ | ✓ | ✓ |
| 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 | 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 other CoreWeave tools; hosted W&B accounts now sign in through Forge, and Weave's docs, pricing and login moved to CoreWeave domains. CoreWeave now recommends Agent Lens, a separate product in public preview sharing Weave's trace backend, for agent tracing, so check which product a quote covers.
Pydantic, the company behind the Pydantic library and Pydantic AI, builds Logfire and is independent. Its main 2026 change was folding the Pydantic AI Gateway into Logfire in March, with guardrails and spend caps added in July.
Logfire is general observability with AI views on top. Its SDKs wrap OpenTelemetry, any standard OpenTelemetry SDK can send to it, and an agent's model and tool calls appear alongside the service's HTTP requests, database queries, logs and metrics. That suits a team whose agent is one part of a larger web application and who wants one place to debug both.
Weave sits beside model work. For a team that already trains, fine-tunes or tracks experiments on Weights & Biases, Weave keeps LLM traces, evaluations and versioned prompts in the same platform, with prompts linkable to the CoreWeave Registry. It does not trace your database or HTTP layer the way Logfire does.
On evaluation the two are closer. Logfire uses pydantic-evals: datasets defined in code or hosted, experiments compared to a baseline in CI, LLM judges that record their reasoning, and live evaluations that run inside your application at a sample rate you choose. Weave's Evaluation API runs a model over a dataset with scorers and shares leaderboards; its monitors score production traces, though its docs now call built-in monitors a previous approach. For human review, Weave documents annotation queues that route traces to reviewers, while Logfire's run annotations are in beta and its queues are early access.
Logfire's guardrails live in its AI gateway. Model calls routed through it are scanned for secrets, personal data and policy violations, with per-rule actions to observe, flag, redact or block, and spend caps stop requests once a budget is spent. The gateway also handles routing across providers with failover and load balancing. Calls that bypass the gateway, and tool actions, are not covered.
Weave's guardrails are scorers you call from Python application code, which can block or modify inputs and outputs before they reach users. There is no gateway and no TypeScript support for guardrails. Neither product documents approvals on agent tool actions.
As of 20 October 2026, Logfire counts records: spans, logs and metrics. Personal is free for 10M records a month with one seat and 30-day retention, Team is $49 a month with 10M records and five seats, Growth is $249 a month with unlimited seats, and extra records cost $2 per million on paid plans.
Weave is billed inside CoreWeave Forge by seats, storage and Weave ingestion. Forge Free includes 1 GB of Weave ingestion a month; Forge Pro starts at $60 a month with 1.5 GB, is limited to teams under 50 people, and lists overage at $0.10 per MB. Logfire bills on the number of records, Weave on their size. See Logfire's pricing and CoreWeave Forge pricing.
Language support follows a Python-first pattern on both sides, with differences at the edges. Logfire ships JavaScript, TypeScript and Rust SDKs and accepts data from any OpenTelemetry SDK, though pydantic-evals is Python. Weave's TypeScript SDK has no class-based scorers, guardrails or cost tracking. For regulated workloads, Logfire's Growth plan includes a boilerplate HIPAA BAA, and Weave offers a HIPAA option on Enterprise.
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. Logfire and Weave both check model inputs and outputs, through a gateway or in code, but neither covers tool actions or approvals. Prefactor has no prompt playground or prompt registry, does not trace your wider application, 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.