LITELLM · FIXED DEMO

See the same tool proposal in three Guard modes.

This fixed example reconstructs a real release test: LiteLLM proposed record_color(color="blue") against a schema requiring at least 1,000 characters. Switch modes to see what the configured policy changes.

Recorded scenario · no live model call · no account required

FIXED REQUEST · openai/gpt-4o-mini
litellm.completion(
    model="openai/gpt-4o-mini",
    messages=messages,
    tools=[record_color_schema],
)

proposal: record_color(color="blue")
schema: color minLength = 1000

GUARD DECISION

No Guard decision

HOST DISPATCH

Host executes @st.tool

The model proposes record_color(color="blue"). The decorated Python function runs and creates a tool span.

Run it with your own agent.

Install the Python SDK, set your Staso and upstream keys, and configure a scoped Guard policy.