Tool gating with AI SDK 7

Short answer: use prepareStep and activeTools from AI SDK 7. reactive-fsm adds a small state policy and a tool budget on top of those primitives.

The problem

A model can call a tool that does not make sense for the current step: asking for a payment before collecting identity, browsing after checkout, or repeating the same lookup because it does not know the flow advanced.

The model only knows the tools you advertise. If a tool is not in the request, the model is much less likely to call it.

The policy

For the current state, reactive-fsm advertises a tool only if all three are true:

1. It appears in config.tools[currentState].
2. Its registry.gates includes the current state.
3. Its condition, if present, returns true.

config.tools is the authority. The registry cannot add a tool that the state does not allow.

AI SDK 7 integration

import { generateText, isStepCount, jsonSchema } from 'ai'
import { openai } from '@ai-sdk/openai'
import { createFSM, type ToolEntry } from 'reactive-fsm'

import { createVercelAdapter } from 'reactive-fsm/adapters/vercel-ai'

interface Ctx {
  contactId: string
}

const fsm = createFSM({
  initialState: 'IDENTITY',
  states: ['IDENTITY', 'BOOKING'] as const,
  tools: {
    IDENTITY: ['validate_dni'],
    BOOKING: ['check_availability'],
  },
})

const registry: ToolEntry<Ctx>[] = [
  {
    name: 'validate_dni',
    gates: ['IDENTITY'],
    build: () => ({
      description: 'Validate the customer DNI',
      inputSchema: jsonSchema({
        type: 'object',
        properties: { dni: { type: 'string' } },
        required: ['dni'],
      }),
      execute: async ({ dni }: { dni: string }) => {
        if (!dni) throw new Error('DNI is required')
        fsm.transitionTo('BOOKING')
        return { valid: true }
      },
    }),
  },
  {
    name: 'check_availability',
    gates: ['BOOKING'],
    build: () => ({
      description: 'Check available slots',
      inputSchema: jsonSchema({ type: 'object', properties: {} }),
      execute: async () => ({ slots: ['10:00', '11:00'] }),
    }),
  },
]

const adapter = createVercelAdapter(fsm, registry, { contactId: 'c-1' })

const result = await generateText({
  model: openai('gpt-4o'),
  prompt: 'Hola, quiero reservar una cita',
  ...adapter.inject(),
  stopWhen: isStepCount(5),
})

What the adapter does

• inject().tools contains the full ToolSet for the run.
• prepareStep returns activeTools with only the tools allowed by the current state.
• The adapter does not mutate the tools object.
• When the state changes during a tool execution, the next step activates the new state's tools.

This is why the ToolSet must contain tools from future states: AI SDK's activeTools can only enable a tool that already exists in the set.

Runnable example

examples/ai-sdk-7 runs the flow with MockLanguageModelV3 and no API key.

Next

• Tool-call budgets
• Native AI SDK vs reactive-fsm