W&B Weave

Weights & Biases Weave is an observability platform for tracking and evaluating LLM applications.

Step 1: Get your W&B API key

In W&B, go to your User Settings and copy your API key.

Step 2: Enable Broadcast in OpenRouter

Go to Settings > Broadcast and toggle Enable Broadcast.

Step 3: Configure W&B Weave

Click the edit icon next to W&B Weave and enter:

• Api Key: Your W&B API key
• Entity: Your W&B username or team name
• Project: The project name where traces will be logged
• Base Url (optional): Default is https://trace.wandb.ai

Step 4: Test and save

Click Test Connection to verify the setup. The configuration only saves if the test passes.

Step 5: Send a test trace

Make an API request through OpenRouter and view the trace in W&B Weave.

Custom Metadata

W&B Weave supports custom attributes and structured inputs for organizing and analyzing your LLM calls.

Supported Metadata Keys

Example

1
{
2
  "model": "openai/gpt-4o",
3
  "messages": [{ "role": "user", "content": "Write a poem about AI..." }],
4
  "user": "user_12345",
5
  "session_id": "session_abc",
6
  "trace": {
7
    "trace_name": "Creative Writing Agent",
8
    "prompt_template": "poem_v2",
9
    "experiment_name": "creative_benchmark",
10
    "dataset_version": "1.0.0"
11
  }
12
}

Attributes and Inputs

Weave organizes trace data into:

• Attributes: Metadata about the call (user IDs, organization IDs, trace identifiers, custom metadata)
• Inputs: The actual request data including messages, model parameters (temperature, max_tokens, etc.)
• Summary: Token usage, costs, and timing metrics

Additional Context

• The user field maps to user_id in attributes
• The session_id field maps to session_id in attributes
• Custom metadata keys from trace are merged into the call’s attributes
• Model parameters (temperature, max_tokens, top_p) are included in inputs for easy filtering

Privacy Mode

When Privacy Mode is enabled for this destination, prompt and completion content is excluded from traces. All other trace data — token usage, costs, timing, model information, and custom metadata — is still sent normally. See Privacy Mode for details.