Knowledge search
A helper that only looks things up — so answers stay grounded without cluttering the main prompt.
How it works
Knowledge search, datastore write, product lookup, email draft — one clear responsibility each.
Give that subagent only the knowledge, tables, or connected apps it needs.
Your main agent calls the right helper when the conversation needs that action.
The same datastore writer can serve support and sales agents without rebuilding it.
Capabilities
Anything that should be accurate, repeatable, and separately tunable.
A helper that only looks things up — so answers stay grounded without cluttering the main prompt.
Create leads, tickets, and feedback with the exact fields you defined.
Draft email, look up Shopify products, update a CRM — one job, clear permissions.
A special kind of subagent that runs after the chat ends — extract fields and take follow-up actions.
Keep the main conversation lean; send heavy tool work to a helper built for that task.
Instructions, tools, and output shape — configured in the builder, not a framework.
In practice
One mega-agent that can do everything usually does nothing reliably.
The visitor-facing agent stays focused on talking; helpers handle the mechanics.
Only the email helper can touch email. Only the datastore helper can write records.
Improve product lookup without touching your support tone — and vice versa.
One well-built helper serves every agent that needs that skill.
FAQ
No. Start with a simple agent and knowledgebase. Add focused helpers when you want reliable writes, lookups, or follow-ups.
Yes. Conversation intelligence uses a subagent that runs after the session ends — with its own fields and optional actions.
Yes. Bind the helpers that match that agent's job — support might get knowledge + tickets; sales might get leads + email.