Weave
Weights & BiasesToolkit for developing and monitoring AI applications from Weights & Biases, providing tracing, evaluation and dataset management for LLM applications. Functions are traced with a decorator, and it builds on the same platform as W&B's experiment tracking. The published package requires a Weights & Biases account to send traces to.
Openness
2 high confidence- license
- Apache-2.0(OSI, SDK/client repo wandb/weave)
- source
- partial(the Apache-2.0 repo publishes the SDK and a ClickHouse-backed trace-server library, but its own trace_server/README.md places the HTTP service that fronts the trace server in W&B's closed core repository)
- backend
- commercial(W&B SaaS platform hosts traces/evals
- eval+tracing-client
- OSS
Weave's wandb/weave repository is Apache-2.0 and publishes more than an SDK, including a working ClickHouse trace-server library, but the HTTP service that fronts it lives in W&B's closed core repository. Using the package requires a Weights & Biases account, and a self-managed install needs a Weave-enabled license from W&B, so it cannot be run from what ships here.
- https://docs.wandb.ai/guides/hosting/ recorded 2026-09-29
The W&B hosting page says "Some features and functionality require an" Enterprise license and compares "deployment options including Multi-tenant Cloud, Dedicated Cloud, and Self-Managed installations".
- https://docs.wandb.ai/weave/guides/platform/weave-self-managed recorded 2026-09-29
The Weave self-managed page lists the prerequisite "Weave-enabled license from W&B Support" and says the W&B Platform is already installed and running.
- https://github.com/wandb/weave recorded 2026-06-04
Apache-2.0 license on the Weave SDK repo; LLM logging/tracing/eval toolkit
- https://raw.githubusercontent.com/wandb/weave/master/LICENSE recorded 2026-09-29
The repository LICENSE file is the Apache License, "Version 2.0, January 2004", the full standard text.
- https://raw.githubusercontent.com/wandb/weave/master/README.md recorded 2026-09-29
The README lists prerequisites of "Python 3.10 or higher" and "A [Weights & Biases account](https://wandb.ai/signup) (free tier available)", and quick start is pip install then weave.init.
- https://raw.githubusercontent.com/wandb/weave/master/weave/trace_server/README.md recorded 2026-09-29
The trace server README diagram places the web service in the closed core repo, labeled "participant TraceWebServer as (in core) trace_server.py" beside the shipped ClickHouseTraceServer.
Adoption
3 high confidenceAdoption is measured by monthly PyPI downloads of the weave package. Weave rides the established Weights & Biases install base, so growth reflects ML teams adding LLM observability inside an existing W&B account as much as it reflects independent adoption of Weave itself.
- https://pypistats.org/api/packages/weave/recent recorded 2026-10-07
817,064 downloads in the trailing 30 days for weave (pypistats last_month)
Capability
4 medium confidenceWeave provides end-to-end tracing and evaluation with LLM judges and custom scorers, plus versioning, feedback collection and production monitoring. Prompt management is lighter than Langfuse's, which keeps the product level with Langfuse rather than ahead of it.
- https://docs.wandb.ai/weave recorded 2026-09-29
Weave docs home: "W&B Weave is an observability and evaluation platform that helps you track, evaluate, and improve your agents and LLM applications." with tracing of calls, "LLM judges and custom scorers", "Version your work", "Monitor and collect feedback" and "Pick a built-in integration for your agent SDK or harness".
- https://github.com/wandb/weave recorded 2026-06-04
logging/debugging LLM traces + rigorous evaluations feature set
Verified 2026-09-29, except some scores, still awaiting re-confirmation