# W&B Weave review

> W&B Weave is a toolkit from Weights & Biases for tracing and evaluating AI agents and LLM applications.

Devtoolwatch Board, the week of Oct 25–31, 2026. Canonical page: https://index.devtoolwatch.com/tools/weave/

## TL;DR

W&B Weave is an observability and evaluation toolkit for agents and LLM applications, built by Weights & Biases. W&B Weave is for Python and TypeScript developers who want traces of conversations, LLM calls and tool calls, plus evaluations with LLM judges and custom scorers. Supported agent SDKs are autopatched after a single weave.init() call.

## W&B Weave at a glance

- **Category:** [Evals and observability](https://index.devtoolwatch.com/categories/evals-and-observability/)
- **Maintained by:** Weights & Biases (organization, [@wandb](https://github.com/wandb) on GitHub)
- **License:** Apache 2.0
- **Language:** Python, TypeScript, JavaScript
- **Repository:** [github.com/wandb/weave](https://github.com/wandb/weave)
- **Website:** [wandb.me/weave](https://wandb.me/weave)
- **npm package:** [weave](https://www.npmjs.com/package/weave) 0.16.9
- **PyPI package:** [weave](https://pypi.org/project/weave/) 0.53.12
- **Latest release:** [v0.53.12](https://github.com/wandb/weave/releases/tag/v0.53.12) on GitHub, Oct 13, 2026
- **First published:** Jun 7, 2023

## What W&B Weave does

- W&B Weave traces agent conversations, turns, LLM calls and tool calls, which appear in the Agents tab.
- W&B Weave traces individual functions with the @weave.op decorator, and those traces land in the Calls tab.
- W&B Weave evaluates agent and application responses using LLM judges and custom scorers.
- W&B Weave autopatches the OpenAI Agents SDK, Claude Agent SDK and Google ADK after weave.init().
- W&B Weave auto-traces plain openai calls into the Calls tab after weave.init().
- W&B Weave accepts OpenTelemetry spans at its OTLP endpoint, with no Weave SDK installation required.
- W&B Weave offers a Conversation SDK with start_conversation, start_turn, start_llm and start_tool for wrapping custom agent loops.

### How it works

You install the Python or TypeScript library and call weave.init() with your team and project name. weave.init() then prints a project URL. Supported agent SDKs and plain openai calls are patched automatically. Custom loops use the Conversation SDK context managers. Traces are sent to the Weights & Biases trace server and viewed in the Agents or Calls tab. OTel-instrumented agents can send spans directly to the OTLP endpoint.

## How to install W&B Weave

**With pip**

```
pip install weave
```

**With npm**

```
npm install weave
```

**With pip, for the OpenAI Agents SDK example**

```
pip install weave openai-agents requests
```

### Requirements

- Python 3.10 or higher
- A Weights & Biases account (free tier available)
- A W&B API key set in WANDB_API_KEY
- An OpenAI API key in OPENAI_API_KEY for the README examples

### First steps

1. Create a Weights & Biases account and set WANDB_API_KEY to the key from wandb.ai/authorize.
2. Call weave.init("<your-team>/<your-project-name>") at the start of your code.
3. Run your agent with a supported SDK, or decorate functions with @weave.op.
4. Open the printed project URL and check the Agents tab or the Calls tab.

## Who W&B Weave is for

- Developers building agents with the OpenAI Agents SDK, Claude Agent SDK or Google ADK who want traces
- Teams evaluating LLM applications, including RAG apps, with LLM judges and custom scorers
- Developers with OpenTelemetry-instrumented agents who want to send spans without adding an SDK

### Limitations

- W&B Weave requires a Weights & Biases account and API key to store traces.
- W&B Weave does not infer provider_name from the model name in start_llm, so you must always pass it.
- W&B Weave's start_session and Session are deprecated and should not be used in new code.
- W&B Weave's engine and boards code is paused while the project focuses on tracing and evaluations.

## Usage data (the week of Oct 25–31, 2026)

- Not ranked this week
- New GitHub stars per week (last 12 weeks): Aug 9: 5; Aug 16: 3; Aug 23: 1; Aug 30: 3; Sep 6: 3; Sep 13: 1; Sep 20: 2; Sep 27: 4; Oct 4: 4; Oct 11: 3; Oct 18: 0; Oct 25: 3

## W&B Weave alternatives

| Rank | Tool | Stars | Downloads/wk |
|---|---|---|---|
| #1 | [SkillSpector](https://index.devtoolwatch.com/tools/nvidia-skillspector/) | 23.7k | – |
| #2 | [Judgeval](https://index.devtoolwatch.com/tools/judgmentlabs-judgeval/) | 1.1k | 80.2k |
| #3 | [A.I.G (AI-Infra-Guard)](https://index.devtoolwatch.com/tools/tencent-ai-infra-guard/) | 7.2k | – |
| #4 | [Langfuse](https://index.devtoolwatch.com/tools/langfuse/) | 36.8k | 9.8M |
| #5 | [Opik](https://index.devtoolwatch.com/tools/opik/) | 23.1k | 709.1k |
| #6 | [promptfoo](https://index.devtoolwatch.com/tools/promptfoo/) | 26.7k | 1.2M |

## W&B Weave FAQ

### Is W&B Weave free and open source?

Yes. W&B Weave's source code is on GitHub (wandb/weave) under the Apache 2.0 license, so you can use it for free.

### How do I install W&B Weave?

Run `pip install weave`. The install section above has the other options from the project's README.

### Is W&B Weave still maintained?

Yes. The wandb/weave repository was last updated on Oct 22, 2026, and the latest release, v0.53.12 on GitHub, came out on Oct 13, 2026.

### How popular is W&B Weave?

W&B Weave has 1.1k GitHub stars as of Nov 1, 2026. It gained 3 new stars in the week of Oct 25–31, 2026, against a 12-week average of 3 a week.

### What are the best alternatives to W&B Weave?

Other LLM evals and observability tools on the Devtoolwatch Board in the week of Oct 25–31, 2026, by momentum: SkillSpector (#1), Judgeval (#2), A.I.G (AI-Infra-Guard) (#3), Langfuse (#4), Opik (#5). The alternatives table above has their numbers.

### What is W&B Weave?

W&B Weave is an observability and evaluation platform from Weights & Biases for agents and LLM applications. W&B Weave traces conversations, LLM calls and tool calls. It also evaluates responses with LLM judges and custom scorers.

### What is the difference between the Agents tab and the Calls tab?

W&B Weave sends agent SDK and Conversation SDK traces to the Agents tab. Functions traced with @weave.op and plain openai calls go to the Calls tab. Use @weave.op for evaluations, scorers and LLM calls that are not an agent loop.

### Can I trace an agent that doesn't use a supported SDK?

Yes. W&B Weave lets you wrap your own loop with start_conversation, start_turn, start_llm and start_tool. In Python these are context managers that close on exceptions.

### Do I need the Weave SDK to send traces?

No. W&B Weave can receive OpenTelemetry spans at its OTLP endpoint. No Weave SDK installation is required for OTel-instrumented agents.

### How does W&B Weave relate to CoreWeave Agent Lens?

W&B Weave's agent tracing workflows are also available in CoreWeave Agent Lens, which the docs call the successor to Weights & Biases Weave. Both services already receive your agent tracing data, so there is nothing to migrate.

## Sources

1. [wandb/weave on GitHub](https://github.com/wandb/weave)
2. [W&B Weave README](https://github.com/wandb/weave/blob/master/README.md)
3. [weave on npm](https://www.npmjs.com/package/weave)
4. [weave on PyPI](https://pypi.org/project/weave/)
5. [W&B Weave website](https://wandb.me/weave)
6. [Release v0.53.12 (Oct 13, 2026)](https://github.com/wandb/weave/releases/tag/v0.53.12)
7. [Release v0.53.11 (Oct 13, 2026)](https://github.com/wandb/weave/releases/tag/v0.53.11)
8. [Release v0.53.10 (Oct 13, 2026)](https://github.com/wandb/weave/releases/tag/v0.53.10)
9. [Hacker News: Whitepaper: Current Best Practices for Training LLMs from Scratch](https://news.ycombinator.com/item?id=35560344)
10. [Hacker News: Weave: Interactive Data Exploration Toolkit by Weights and Biases](https://news.ycombinator.com/item?id=36237307)
11. [Devtoolwatch Board, the week of Oct 25–31, 2026](https://index.devtoolwatch.com/)