Interface: ToolConfig<S>

Defined in: packages/sdk/src/renderer/templates/chat-tools.ts:269

A tool the model can call. Matches the Vercel AI SDK's tool() — inputSchema + execute(input, options) — with confirm and the reply / ui / plan() returns added on top. A vanilla AI-SDK tool's bare-value return is treated as something for the AI to phrase.

Type Parameters

Type Parameter	Description
S extends z.ZodType	The Zod schema for the tool's input.

Properties

Property	Type	Description	Defined in
description	string	What the model reads to decide when to call this tool.	packages/sdk/src/renderer/templates/chat-tools.ts:271
execute	(input: output<S>, options: ToolRunOptions) => | ChatToolReturn | Promise<ChatToolReturn>	The tool's action — runs skill-side (Node), so fetch, secrets, and (via closure) ctx are available; the second argument carries the runtime agent and the raw text. Return reply for verbatim text, a bare value to let the AI phrase it, ui for rendered blocks, or plan(). Throwing is safe: it is caught and shown as a graceful error.	packages/sdk/src/renderer/templates/chat-tools.ts:279
inputSchema?	S	Zod schema for the model-supplied input.	packages/sdk/src/renderer/templates/chat-tools.ts:284
confirm?	(input: output<S>, options: ToolRunOptions) => ChatConfirmResult | Promise<ChatConfirmResult>	Require human approval before execute runs. Return a prompt (string or full spec) to gate the call — execute fires only after the user approves, on every model path. Return a falsy value to skip the gate for this call (conditional confirmation).	packages/sdk/src/renderer/templates/chat-tools.ts:291