Using AI in Ambassadors

Ambassadors have two built-in AI paths. Choose the one that fits your workload — or use both in the same ambassador for different tasks.

One AI — Full-Power AI

One AI routes your request through the One intelligence layer, which selects the best model for the task. It supports all available models, compliance-aware routing, and the full range of AI capabilities.

When to use

• Complex reasoning, planning, or strategic analysis
• High-quality writing (reports, summaries, proposals)
• HIPAA-compliant data processing (regulated industries)
• Tasks where quality matters more than speed

Code example

export default async function handler(request, ctx) {
  const result = await ctx.ai.generate({
    prompt: 'Analyse this customer feedback and identify the top 3 themes...',
  })

  return new Response(result.text)
}

Model selection

One automatically selects the best model for your task based on complexity, cost, and compliance requirements. You can override this by specifying a model:

const result = await ctx.ai.generate({
  prompt: 'Write a detailed market analysis...',
  model: 'gpt-4o',
  maxTokens: 4000,
})

HIPAA compliance

If your organization's workspace is classified as HIPAA-regulated, One automatically routes all AI requests to HIPAA-compliant models. No code changes needed — the routing is handled by the platform.

---

Edge AI — Fast AI

Edge AI runs an edge-optimised model directly on the same global network as your ambassador. There's no round-trip to an external service — the AI runs alongside your code, making it very fast and very affordable.

When to use

• Data extraction from text, forms, or images
• Classification, categorisation, and tagging
• Summarisation at high volume
• Structured output (get JSON back, not free text)
• Multi-turn conversations where speed matters

Text generation

export default async function handler(request, ctx) {
  const result = await ctx.ai.edge.generate({
    prompt: 'Extract the company name and industry from this text: ' + inputText,
  })

  return new Response(result.text)
}

Structured output

Get validated JSON back instead of free text. Provide a schema describing the shape you expect:

const result = await ctx.ai.edge.structured({
  prompt: 'Extract the following from this invoice: ' + invoiceText,
  schema: {
    company: 'string',
    total: 'number',
    currency: 'string',
    date: 'string',
  },
})

// result.data.company, result.data.total, etc. are typed
await ctx.tables.insert('invoices', result.data)

Vision (processing images)

Edge AI can process images alongside text prompts — useful for receipts, documents, photos, and screenshots:

const imageBuffer = await request.arrayBuffer()

const result = await ctx.ai.edge.vision({
  prompt: 'Describe what you see in this image',
  images: [imageBuffer],
})

Conversation caching

For multi-turn conversations, pass a sessionId to keep the conversation context warm. Each subsequent turn is faster and cheaper because the model remembers the previous context:

const sessionId = crypto.randomUUID()

const turn1 = await ctx.ai.edge.generate({
  prompt: 'You are a customer support agent. The customer says: ' + message,
  sessionId,
})

// Later, in a follow-up request with the same sessionId:
const turn2 = await ctx.ai.edge.generate({
  prompt: 'The customer replies: ' + followUp,
  sessionId,
})
// turn2 is faster and cheaper — the context is already cached

Limitations

• Not HIPAA-compliant — if your workspace requires HIPAA compliance, Edge AI will return an error directing you to use One AI instead
• Single model — Edge AI uses an edge-optimised model. You cannot select a specific model like you can with One AI
• No streaming — Edge AI returns the complete response. For streamed responses, use One AI

---

Choosing the Right Path

Use this table to decide which path fits each task in your ambassador:

Task	Recommended path	Why
Extract fields from a form submission	Edge AI (structured)	Fast, cheap, returns typed JSON
Write a detailed report	One AI	Needs premium reasoning and writing quality
Classify incoming support tickets	Edge AI	High volume, simple categorisation
Process uploaded receipts	Edge AI (vision)	Image processing at low cost
Analyse financial data for compliance	One AI	May involve regulated data, needs precision
Summarise meeting notes	Edge AI	Straightforward summarisation
Generate a strategic recommendation	One AI	Complex reasoning required
Qualify leads from form data	Edge AI (structured)	Extract and categorise at volume

You can use both paths in the same ambassador — for example, use Edge AI to extract data from a document, then pass the extracted data to One AI for deeper analysis.

---

Credit Usage

Both AI paths consume credits from your organization's WaymakerOne Pass allocation:

• One AI credits are tracked in real time as part of the request
• Edge AI credits are tracked asynchronously — they appear in your usage within a few seconds

View your AI credit usage in the Invocation Logs for per-ambassador tracking, or in your organization's billing dashboard for overall usage.

---

Next steps

• Creating Ambassadors — set up a new ambassador
• Invoking & Logs — run ambassadors and view logs
• Serverless Functions — ambassador overview