> ## Documentation Index > Fetch the complete documentation index at: https://dataleap-469bb19d-docs-from-request-2026-08-21-templates.mintlify.site/llms.txt > Use this file to discover all available pages before exploring further. # Sub-agents & model tiers > What sub-agents are (Dataleap handles them for you) — and how to pick the right model tier. An agent isn't always just one agent — behind the scenes it can be a whole system: a main agent delegating parts of the job to **sub-agents**, running as one logical process. The important part: **this happens automatically**. Dataleap decides when a workflow benefits from sub-agents and architects them for you. There's nothing to set up, nothing to manage — this page just explains the concept so you recognize it when you see it. ## What a sub-agent system looks like Diagram: a main agent delegating to three sub-agents — research, copywriting, and a scraper called fifty times Diagram: a main agent delegating to three sub-agents — research, copywriting, and a scraper called fifty times The main agent breaks the goal apart, delegates to sub-agents, and merges the results. Dataleap typically reaches for this when a workflow involves **batch processing** (the same task run many times), **heavy intermediate data** (a sub-agent digs through it and returns just a summary, so the main agent stays lean), or **distinct phases** that each need their own focused instructions. If your agent uses sub-agents, they appear under its **Tools** — you can look at each one, and that's all you ever need to do with them: The Tools tab of an agent showing one sub-agent, the Candidate Scorer A sub-agent belongs to its parent agent; for logic that several agents should share, that's what [skills](/scale/skills) are for. ## Model tiers What you *can* adjust: the **preset** — separately for the main workflow agent and for each sub-agent. Higher tiers are more capable and more expensive. The model selector in the chat bar showing Basic, Pro, Expert, and Customize While the agent takes shape, prefer **Expert** — building benefits from the best judgment. Setup conversations can't use the **Basic** preset. Execution agents default to **Pro**. That's the right setting for \~90% of use cases — you rarely need to touch it. Two paths, depending on what you observe: * **The agent runs perfectly fine?** Try lowering the execution agent's preset — same output, cheaper runs. * **The agent doesn't work right** — information goes missing, or runs time out? Level it up toward **Expert** and test again. You set the preset with the model selector — in the chat bar for the setup conversation, in the **Workflow** tab for the main agent, and in a sub-agent's **Tools** tab. **Customize** lets you pick a specific model and a **Reasoning Level** instead of a preset. Which models appear depends on what an admin has enabled on **Models**. Admins also choose which model each preset uses — see [Models](/admin/models). Complex agents that pull from many sources may genuinely need a higher tier — a smaller context can't hold everything the workflow touches. And if the workflow is **data-heavy and deterministic**, the better fix is often [programmatic tool calling](/scale/programmatic-tool-calling) instead of a bigger model. See how sub-agents and model tiers work in the product. Next: share behavior across agents the right way. How tier choice fits into overall spend control.