Managing tools
Open AI agents → edit agent → Tools, or go to Integrations → Tools for workspace-wide toggles. Each tool has:- a key (e.g.
shopify.orders.lookup) - a short description of what it does
- an enabled/disabled switch
- optional config (e.g. which store to query)
Built-in tools
| Tool | What it does | Requires |
|---|---|---|
| Handoff to human | Assign the conversation to a teammate when the AI can’t help | — |
| Tag conversation | Apply a tag the agent picks (e.g. returns) | — |
| Collect email | Ask the visitor for an email if not known | — |
Integration-gated tools
These appear only when the relevant integration is connected:
Connect the integration first, then the tools become switchable on your agent.
When tools fire
The agent decides per-message whether to use a tool. The decision is based on the system prompt and the conversation — e.g. “customer said ‘where is my order #1234’” → callshopify.orders.lookup.
You can force tool usage with the system prompt: “Whenever a customer mentions an order number, call shopify.orders.lookup before answering.”
Seeing tool calls
In any conversation, open Sources on the context drawer — it shows which knowledge passages and which tools were used for each reply. Handy when an answer looks surprising.Tool limits
- Some plans cap how many tools can be active per agent. See Plans.
- Each tool call counts against your credit allowance just like a reply.
- If a tool fails (e.g. Shopify down), the agent falls back to answering from knowledge alone and mentions it couldn’t fetch live data.