Graph view: Strike docs

The Strike console graph view shows one agent run from trace spans exported by strikeflow (POST /v1/traces). It reads the spans strikeflow exports to POST /v1/traces.

What you need

STRIKE_API_KEY with permission to ingest traces
STRIKE_PROJECT set to the same string you pass to strike.trace(..., project=...) or strike.configure(project=...)
Decisions recorded inside a trace (or standalone with run_id)

Python

import strikeflow as strike

strike.configure(project="support-bot")
with strike.trace("support-agent", project="support-bot", run_id="ticket-4411"):
    strike.decide("Refund?", "yes_no", context="Duplicate charge on invoice 4411.")
strike.flush()

How to read the graph

Span `type`	Drawn as
`run`	Root node (run name, latency)
`step` with `metadata.kind` `llm`	LLM call (`metadata.provider`, `metadata.model` on the span)
`step` with `metadata.kind` `tool`	Tool call (`metadata.tool`, optional `metadata.host`)
`step` with `metadata.kind` `node` / `step`	Generic step
`decision`	strike-1 node with options and probabilities

Edges follow parent_id → span_id.

Decision edges

For each option in decision.options, the UI shows a branch labelled with its probability from decision.probabilities.

The branch matching decision.path_taken is drawn as the path your router took (solid).
Other option branches are dashed.
If decision.metadata.fallback.used is true, the chosen answer edge is de-emphasised and an extra edge points to fallback.to with the reason.

Typical fallback reasons from strike_router or manual annotate: low_confidence, error, missing, unmapped_answer.

Where to open it

Spans missing? Check STRIKE_PROJECT, call strike.flush(), and check your API key (Authentication).