Runs

Runs are the platform's execution ledger. They are how you move from "the model returned something surprising" to "here is exactly what executed, with which prompt version, model, tools, and runtime cost."

What runs record

Current run records expose fields such as:

• prompt identity
• prompt version
• provider and model
• transport type
• input and output token counts
• estimated cost
• latency
• completion timestamps
• step counts, including tool calls

Current app views

The Runs page shows the broad execution index for prompt runs. It also computes summary metrics directly from the visible data, including:

• total run count
• average latency
• number of tool-call runs
• total estimated cost

The detail page goes deeper into:

• step timeline
• tool activity
• output/result snapshot
• input snapshot
• model/runtime summary

Run detail

The main operational value of runs is that they preserve composition and execution evidence after the fact. That makes them useful for:

• debugging incorrect outputs
• spotting prompt regressions
• verifying tool usage
• understanding runtime cost and latency

API workflow

curl "$API_BASE/api/runs?organizationId=org_123" \
  -H "Authorization: Bearer $TOKEN"

curl "$API_BASE/api/runs/run_123?organizationId=org_123" \
  -H "Authorization: Bearer $TOKEN"