Answer customer questions from your docs with automatic email escalation for unanswered queries. Cut support ticket volume by 60%. Answers from docs in seconds, escalates what it cannot solve.

This is the whole blueprint, nothing held back: the model, the tools, the exact prompt, the source on disk, and the setup. Everything you need to run Support Bot yourself is on this page, and it deploys in an afternoon.

See it running

Agents like Support Bot run behind a live support console. Here is that pattern, live: every action stays visible and reviewable. Click into it.

crescendo.51ultron.com/kits/support-console

What it does

Support Bot handles the work end to end:

Answer questions from any documentation set automatically
Smart keyword search across docs without reading full files
Cite source URLs for every answer provided
Escalate unanswered questions to your team via Resend email

The anatomy

An agent is a composition, not a prompt: a model, the tools it is allowed to call, the integrations that give those tools reach, and the rules that keep it honest. Here is Support Bot on one card.

Support Botblueprint

Model: claude-sonnet-4-6
Tools: fetch_doc_pagesend_email
Integrations: Resend
Guardrails: /workspace/llms.txt, index of all doc pages with links; /workspace/llms-full.txt, full documentation content

The instructions

Every run is governed by this prompt. It is short on purpose: enough to make Support Bot reliable, not so much that it can misread itself. Copy it as a starting point for any agent you build.

Support Bot

You are a support agent. Your primary job is to answer user questions using the product documentation loaded into your workspace. When you cannot find an answer in the docs, offer to forward the question to the team via email. RULES: 1. Documentation files are available locally: - /workspace/llms.txt — index of all doc pages with links - /workspace/llms-full.txt — full documentation content 2. ALWAYS search local files first using grep. Do NOT read the entire file. 3. Use grep with relevant keywords to find the right sections. 4. When answering, cite the relevant documentation page URLs. 5. Be concise and accurate. Provide code examples from the docs when relevant. ESCALATION — when to use send_email: - You searched the docs thoroughly and could not find a relevant answer. - The user explicitly asks to contact the team or get human help. - The question involves account-specific issues, billing, or bugs. When sending an email, include the full conversation context.

The source, on disk

Three files carry the whole agent: the definition with its tools, the environment it expects, and the setup. Walk them.

support-bot/support-bot.ts1.0 KB

1import { agent, tool, Sandbox } from "@agent-sdk"

2import { z } from "zod"

4export default agent({

5 runtime: "claude-code",

6 model: "claude-sonnet-4-6",

7 permissionMode: "bypassPermissions",

8 maxTurns: 25,

9 systemPrompt: `...`, // see System Prompt section above

10 sandbox: Sandbox({

11 setup: [

12 // fetches docs from DOCS_URL into /workspace/llms.txt

13 ],

14 }),

15 tools: {

16 fetch_doc_page: tool({

17 description: "Fetch a documentation page by URL",

18 inputSchema: z.object({ url: z.string().url() }),

19 execute: async ({ url }) => {

20 const res = await fetch(url)

21 return { content: [{ type: "text", text: await res.text() }] }

22 },

23 }),

24 send_email: tool({

25 description: "Escalate unanswered question via email",

26 inputSchema: z.object({

27 to: z.string().email(),

28 subject: z.string(),

29 body: z.string(),

30 }),

31 execute: async ({ to, subject, body }) => {

32 // POST to Resend API

33 },

34 }),

35 },

36})

The business case

If you run an agency, Support Bot is also a product. It sells into saas, developer tools, fintech, and the numbers work at very small scale.

$3,000

setup you can charge

$500/mo

monthly retainer

60%

less manual work

Ship it

The whole install is a handful of commands and the keys from the env file. Run the list, then point Support Bot at real work.

Deploy checklist

0 of 5

That is the entire blueprint: one model, 2 tools, and a prompt. Built by hand it costs you an afternoon. Inside Ultron it costs you a sentence, because you can build this exact agent in the workforce and watch it run before you commit to anything.

Build Support Bot in Ultron

Describe what you want

Start from this blueprint or your own words. Ultron reads the intent and picks the tools.

It builds the agent

The model, the tools, the prompt, the console. Assembled in front of you, nothing to wire.

Watch it run

Test it live in the chat, then put it to work. You review what it does, it does the rest.