AI Multi-Agent Automation Blueprint

Description

This is the blueprint behind a production multi-agent system running 24/7 since 2025, serving users through a Telegram bot and a web interface, routing each job to the cheapest model that can do it, and modifying its own codebase under approval. You get the runnable starter code and the 53-page guide that explains why it is built that way.

What you get

One ZIP with two things side by side:

• The PDF guide – 53 pages, 16 chapters. Architecture, the seven agents that run in production, model routing with the real cost table, secrets handling with HashiCorp Vault and auto-unseal, the Telegram plus FastAPI dual interface, deployment, monitoring, and a scaling playbook for adding an agent or a provider.
• The starter repository – runnable Python, not snippets. A FastAPI app with /chat and /health endpoints, an intent router, a multi-provider LLM client (Anthropic and OpenAI) with an offline fallback, four agent implementations on a shared base class, a monitoring module, focused system briefs, requirements.txt, .env.example, a systemd unit file, and a health-check script.

Who it is for / who should not buy

Built for Python developers who want a production architecture rather than another tutorial repo, indie hackers who are already paying too much for API calls, and engineers who need to explain a multi-agent design to someone and would rather show a working one.

Not for you if you have never deployed anything: this assumes you can read Python, run uvicorn, and put an API key in a .env file. Not for you if you want no-code – see the AI Automation Playbook instead.

How it works

A FastAPI service receives a request, an intent router decides which specialist agent should handle it, and a multi-provider LLM client sends the job to a model chosen by cost and quality tier rather than always reaching for the most capable (and most expensive) one. Each agent runs on a shared base class with its own focused system brief. Without an API key, the client falls back to an offline stub so the app still starts and you can wire up routing and endpoints before spending anything.

Included agents and chapters

The starter code ships four agents so it runs on the first try:

• Code (CodeSmith) – the self-modifying one. Reads the repository, proposes a patch, and waits for your approval before anything is applied.
• Content – long-form drafting and repurposing, with a draft-then-humanize pipeline.
• E-commerce – product research and listing work.
• General – intent classification and everything that does not need a specialist.

The PDF describes three more agents that run in the author’s own system – orchestrator, auto-promoter, and market monitor – with the architecture for wiring them in, though the starter code itself ships only the four above.

Routing matters because sending everything to the most capable model is how the bill gets to be four figures. The router picks the tier per job: a classification goes to the cheap fast model, a client-facing draft does not. The PDF includes the query-type-to-model table and the monthly cost breakdown for running the whole documented system under $90.

Adding your own agent takes about 30 minutes: subclass the base agent, register it in the router with a keyword rule, restart. No existing agent needs to be touched.

Important limitations

• This is a starter kit and a guide, not a hosted service. You run it on your own machine or server.
• You need your own API keys (Anthropic or OpenAI, or both). API usage costs are separate from this purchase and are billed to you by the provider.
• No guaranteed income or results are implied – this is architecture and working code, not a business-in-a-box.
• The self-modifying Code agent proposes changes only; applying them is gated behind a human-approval step by design, and the guide explains what happens without that step.
• The three additional agents in the PDF (orchestrator, auto-promoter, market monitor) are documented, not shipped as running code here.

Need this running on your own server? Write to admin@creatifystore.com and say where you got stuck – deployment, cost, monitoring, or tests. That message is how this product gets improved, so specifics help more than “it doesn’t work.”

Frequently asked questions

What Python version do I need?

3.10 or newer. Every dependency is listed in requirements.txt.

Do I need paid API keys?

Yes – Anthropic or OpenAI, either works, and both is better because that is what the failover is for. The client falls back to an offline stub when no key is present, so the app still starts without one.

Is this a course or code?

Code plus the document that explains why it is built that way. You can run the starter today and read the reasoning afterwards.

Does the self-modifying agent actually work?

Yes, with a human-approval step in the middle. That step is deliberate, and the guide explains what happens without it.

What if something does not run?

Write to admin@creatifystore.com and paste the error. A person answers within 24 hours.

Can I use this commercially?

Yes. The starter code is yours to use inside your own projects and business, commercial ones included. You may not resell or repackage the files themselves as a product of your own.

AI was used in creating this product and in the system it documents. All of it is human-curated and verified in production.

Questions or a problem with the files? Write to admin@creatifystore.com – a person answers within 24 hours. Refunds: full refund within 14 days if the files are faulty or not as described.

Looking for the no-code side of this business? See the AI Automation Playbook or the AgentOps Playbook. A free starting point is the free automation toolkit, or see automation services if you would rather have this built and run for you. The weekly numbers behind this one-person studio, unflattering ones included, are on the Medium ledger.

Version: 2.2

Delivery: Instant ZIP download (0.4 MB)

Price: US$29 one-time

Support: admin@creatifystore.com