LLM provider SDKs
LLM provider SDKs are the client libraries you use to call a model API directly. Install the CoreWeave Forge SDK and calltracing.init("[YOUR-TEAM]/[YOUR-PROJECT]"), and the Forge SDK patches the provider SDK to record each call. Agent Lens shows only calls that run inside a conversation, so make your calls inside a Conversation, as each integration page shows:
Agent SDKs
Agent SDKs are libraries for building agents and multi-agent workflows in your own application code. These integrations use the Weights & Biases Weave SDK, because the Forge SDK doesn’t support these frameworks yet. Addweave.init("[YOUR-TEAM]/[YOUR-PROJECT]") to your script, and Weave patches the agent SDK. The traces appear in Agent Lens, because Agent Lens and Weave share the same trace data.
Agent SDKs run one turn per call, so weave.init() alone doesn’t group turns into a conversation. Pass the SDK’s own session or group identifier on every turn. Each integration page shows which identifier to use:
Agent harnesses
Agent harnesses are end-user agent runtimes, such as coding agents and developer tools. These integrations rely on Weights & Biases Weave, because Agent Lens doesn’t fully support these yet. Install the Weave plugin or extension for your harness, and then start a session in the harness to begin tracing. The harness owns the session, so the plugin groups turns into conversations for you:Custom agents and OpenTelemetry
If your agent isn’t built with one of the SDKs above, or it already emits OpenTelemetry spans, use the Forge SDK directly or send your spans to Agent Lens. Agent Lens accepts any OTel span and gives special handling to GenAI semantic-convention attributes so spans render in the Conversations tab:- Quickstart: Manually instrument an agent: a walkthrough of a small multi-turn agent.
- Trace your agents: the full reference of agent helpers and concepts.
- Send OpenTelemetry spans to Agent Lens: point an existing OTel pipeline at the Agent Lens endpoint, with no SDK required.