LLM Integrations

One line of code to auto-trace LLM provider calls. No wrappers, no changes to your existing LLM code.

Available

How It Works

Call the patch function. Every API call to that provider automatically emits an LLM span — model, tokens, latency, errors. Both sync and async clients are covered. Streaming is fully supported.

st.integrations.patch_anthropic()
st.integrations.patch_openai()

# Use your LLM clients normally — traces happen automatically

Other Providers

For providers without a built-in integration, use @st.trace(kind="llm"):

@st.trace(name="cohere_call", kind="llm")
def call_cohere(prompt: str) -> str:
    response = cohere.chat(...)
    return response.text

You get timing and error tracking, but not automatic token extraction.