Back to catalog / Agent workflows

Query independent tools in parallel within one analysis turn

Let Claude query independent data sources together and synthesize results, avoiding serial waits.

Goal

Reduce latency for analyses that need multiple independent read-only facts.

Identify independent query tools and define strict input/output schemas. [parallel]
Configure tool_choice={"type":"auto"} and inspect multiple tool_use blocks in one assistant turn. [parallel]
Execute tools concurrently, return a tool_result per tool_use_id, and verify synthesis keeps sources distinct. [parallel]

One turn contains multiple independent read calls; latency beats the serial baseline and results pair by call ID.

If the model chooses only one tool, the notebook suggests a batch meta-tool; still cap concurrency.

Do not parallelize side-effecting tools; cap calls per turn and downstream rate.

Parallelize only independent, idempotent calls; use a transaction or explicit ordering when those guarantees are required.

These values describe the upstream repository, not popularity or likes for an individual skill or playbook.

No metrics snapshot is available yet.

License: MIT