Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.wandb.ai/llms.txt

Use this file to discover all available pages before exploring further.

Weave provides two ways to integrate with your AI stack:
  • Trace agents: For multi-turn agentic applications built with agent SDKs (such as the OpenAI Agents SDK or Google ADK) or run inside agent harnesses (such as Claude Code, Codex, or Pi.dev). These integrations capture sessions, turns, LLM calls, and tool calls, and render them in the Weave Agents view.
  • Trace LLM applications: For code that calls LLM providers (OpenAI, Anthropic, Bedrock, and others) or uses orchestration frameworks (LangChain, LlamaIndex, DSPy, and others). These integrations capture individual LLM calls and pipeline steps as Weave Calls in the Traces view.
If you’re not sure which path to take, start with Trace your agents for new multi-turn agent work, or Trace LLM applications for existing LLM-call workflows.

​
Trace agents

The Weave SDK models the full lifecycle of a multi-turn agent: sessions, turns, LLM calls, and tool calls. For supported agent SDKs and harnesses, Weave autopatches the framework so you only need to call weave.init() — every agent invocation, sub-agent handoff, model call, and tool call is captured automatically and rendered in the Agents view. For custom agents, you can instrument any agent code by hand using weave.start_session, weave.start_turn, weave.start_llm, and weave.start_tool. See the agents quickstart for a walkthrough.

​
Integrate Weave with agent SDKs

Agent SDKs are libraries for building agents and multi-agent workflows in your own application code. Weave autopatches the following SDKs:

​
Integrate Weave with agent harnesses

Agent harnesses are end-user agent runtimes (such as coding agents and developer tools) that produce spans Weave can capture. Install the appropriate plugin or extension and your harness sessions are routed to the Agents view:

​
Build your own

Use the Weave SDK directly to instrument custom agents, including any agent that emits OpenTelemetry spans. Weave accepts any OTel span and has special handling for GenAI semantic-convention attributes so your spans render correctly in the Agents view of the Weave UI. See the the Trace an agent quickstart for information on how to trace custom agents.

​
Trace LLM applications

If your application calls an LLM provider’s API directly or uses an orchestration framework, Weave can automatically intercept traces (using autopatching) for many popular libraries and frameworks. By importing the Weave SDK into your code and initializing with weave.init, each request is recorded as a Weave Call with inputs, outputs, latency, token usage, and cost. For libraries Weave doesn’t autopatch, you can manually apply Weave Ops to your code to capture traces.

​
LLM providers

LLM providers are the vendors that offer access to large language models for generating predictions. Weave integrates with these providers to log and trace the interactions with their APIs: Local Models: For when you’re running models on your own infrastructure.

​
Frameworks

Frameworks help orchestrate the execution pipelines in AI applications. They provide tools and abstractions for building complex workflows. Weave integrates with these frameworks to trace the entire pipeline:

​
RL Frameworks

​
Protocols

Weave integrates with standardized protocols that enable communication between AI applications and their supporting services: