Reading execution traces
Every response Sol gives ends with:
> *[Ver trace completo](/traces/abc-123-...)*
That link is your ground truth. Click it when:
Something doesn't match what Sol said
You want to verify a cost or timing claim
A process failed and you want to know the exact step
You're planning to scale this process and want to see how it actually runs
What you see
Header:
Process name + status (completed / failed / aborted)
Total duration
Steps completed / total
AGAS tokens used + USD cost
Retry count (if any)
Error message at the top if it failed
Steps section: each step is an expandable card showing:
Kind (capability / foreach / analyze / output / transform)
ok/error badge
Duration
Input (click to expand)
Output (click to expand) — JSON of what the step produced
Capability executions section: for every capability the process invoked, a sub-card with:
Capability name + description
Mode (replay vs explore_fresh vs explore_repair)
Vision tokens and LLM tokens used
Input and output JSON
Costs section: itemized table of every billable event:
LLM tokens
Vision tokens
Email sends
Storage (if applicable)
Why traces matter
Sol has defensive mechanisms against hallucination (the response-verifier redacts unsupported claims). But the definitive answer always lives in the trace.
If Sol ever says something that feels off, the trace tells you: is this real? Does the data back it up? Is the cost what Sol claims?
Sharing a trace
Trace URLs are per-tenant: only users who own the business can access their traces. Internal staff with service role can also read any trace for support.
For privacy: share a trace by copying the URL to your own team. External parties (investors, auditors) need a staff-generated summary; we don't expose direct trace URLs to third parties.