A standard run should start executing in under two seconds. If yours are slower, the trace will tell you why — latency in mynd is measurable, not mysterious.
Read the trace first
y0 trace <run-id> --timings
# context.load 4,210ms ← the problem is usually here
# model.route 310ms
# step.execute 1,840ms
# total 6,360msThe usual suspects
- Oversized context — loading 'everything' is slow and rarely helps the agent. Narrow the context to what the task needs.
- Cold connectors — a tool that has not synced recently fetches live, which is slower. Frequent contexts stay warm automatically.
- Upstream tool latency — if your CRM takes 3s to answer, the run inherits it. The trace attributes time per connector so you know whom to blame.
- Region mismatch — a workspace pinned to eu-central calling from Sydney pays the round trip. Check your residency setting.
When it is us
If trace timings show time lost inside model.route or step.execute with normal inputs, that is our side. Check status.myndlabs.ai first, then write to support with the run ID — a run ID lets us see the exact execution and answer with specifics instead of sympathy.