Platform
W&B Models
W&B Weave
W&B Inference
W&B Training
Release NotesTracing and observability
Track every call in your LLM application to understand behavior, debug issues, and optimize performance.| Feature | Description |
|---|---|
| Automatic LLM tracing | Weave automatically traces calls to supported LLM providers (OpenAI, Anthropic, Google, and others) without code changes. |
| Custom function tracing | Use @weave.op() to trace any function and capture inputs, outputs, timing, and errors. |
| Nested call tracking | See parent-child relationships between calls to understand your application’s execution flow. |
| Cost tracking | Automatically calculate token usage and costs for LLM calls based on provider pricing. |
| Rich media support | Log and display images, audio, video, and other media types in traces. |
| Trace comparison | Compare traces side-by-side to understand differences between executions. |
| Thread tracking | Group related calls into threads for multi-turn conversation analysis. |
| OpenTelemetry support | Send traces from OpenTelemetry-instrumented applications to Weave. |
Evaluation
Systematically test your LLM application against curated examples to measure and improve quality.| Feature | Description |
|---|---|
| Evaluation pipelines | Run your application against datasets and score outputs with custom or built-in scorers. |
| Datasets | Create, version, and manage collections of test examples. |
| Custom scorers | Write scoring functions tailored to your application’s quality criteria. |
| Built-in scorers | Use pre-built scorers for common tasks like hallucination detection, summarization quality, and more. |
| LLM judges | Use LLMs to evaluate outputs with customizable judge prompts. |
| Leaderboards | Compare model performance across evaluations with configurable leaderboard views. |
| EvaluationLogger | Log evaluation results from external systems or custom evaluation workflows. |
Versioning
Track changes to your prompts, models, and data to understand what changed and reproduce results.| Feature | Description |
|---|---|
| Object versioning | Weave automatically versions any tracked object when it changes. |
| Model tracking | Version model configurations, parameters, and code together. |
| Prompt management | Version prompt templates and track which prompts were used in each call. |
| Dataset versioning | Track changes to evaluation datasets over time. |
| Op versioning | Automatically version functions when their code changes. |
Production tools
Collect feedback, monitor quality, and protect your application in production.| Feature | Description |
|---|---|
| Feedback collection | Capture user feedback, annotations, and corrections on production traces. |
| Guardrails | Run scorers on production traffic to catch issues before they reach users. |
| Monitors | Track quality metrics over time and set up alerts for regressions. |
| PII redaction | Automatically redact sensitive information from traces. |
| Attributes | Tag traces with custom metadata for filtering and analysis. |
Developer experience
Tools to help you iterate faster during development.| Feature | Description |
|---|---|
| Playground | Test prompts and models interactively with the Weave Playground. |
| Evaluation playground | Run quick evaluations on traced calls directly from the UI. |
| Saved views | Save and share filtered views of your traces and evaluations. |
| Comparison tools | Compare objects, traces, and evaluation results side-by-side. |
Integrations
Connect Weave to your existing tools and infrastructure.LLM providers
Weave automatically traces calls to these providers:- Amazon Bedrock
- Anthropic
- Azure OpenAI
- Cerebras
- Cohere
- Google (Gemini, Vertex AI)
- Groq
- Hugging Face
- LiteLLM
- Mistral
- NVIDIA NIM
- OpenAI
- OpenRouter
- Together AI
- Local models (Ollama, vLLM)
Frameworks
Weave integrates with popular LLM frameworks:- OpenAI Agents SDK
- LangChain
- LlamaIndex
- DSPy
- Instructor
- CrewAI
- Smolagents
- Pydantic AI
- Google ADK
- Agno
- Koog
- AutoGen
- Verdict
- Verifiers
Protocols
- Model Context Protocol (MCP)
Platform and deployment
| Feature | Description |
|---|---|
| W&B Cloud | Use Weave on W&B’s managed cloud infrastructure. |
| Self-managed deployment | Deploy Weave on your own infrastructure for data residency requirements. |
| Team collaboration | Share projects, traces, and evaluations with your team. |
| Access control | Manage permissions with W&B’s identity and access management. |
SDKs
Weave provides official SDKs for:- Python: Full-featured SDK with all Weave capabilities
- TypeScript: SDK for Node.js applications with core tracing and evaluation features