Skip to main content
Every run has an effort level. It controls how many angles the task is broken into, how many sources the agent reads, how many research rounds it runs, and how long it can take. Set a default effort on the agent, and override it per run when needed.

Effort levels

low runs a single fast pass: search, read the best results, answer. Each level up widens the fan-out, adds research rounds, and raises the bar on source verification. high is the default.

Pricing

Each level has one flat price. Billing is runs × effort price. Compute, retrieval, and storage are included. Two guidelines:
  • Match effort to the use case. Free-text research runs at medium. Enrichment, dataset building, and research with an output schema run at high. x-high is for hard enrichment fields and exhaustive single-site catalogs.
  • Right-size on outcomes. Compare rows returned, cells filled, and a spot check of citations between tiers on the same inputs. The confidence grade alone does not tell you whether a higher tier is needed.

Choosing Effort and Budget

Recommended tier per use case, what each tier buys, and how to size a job

Set the effort

The agent carries a default. Any run can override it:
Templates ship with a tuned default. company-profile runs at medium; competitive-intelligence and due-diligence run at x-high.

Max

Coming soon. max removes the ceiling: you set the budget per run, for 30-minute to hours-long tasks with the widest fan-out and deepest verification. For workloads that need it, contact Nimble.

Next Steps

Quickstart

Run your first task at the default effort

Trust

Use confidence grades to right-size effort

Use Cases

Research, Enrichment, and Dataset Building