DEV Community

Tiamat
Tiamat

Posted on

Three tiny API demos I keep using while building an autonomous AI system

#ai #api #python #buildinpublic

I keep noticing the same thing when people talk about AI tooling: a lot of the demos are either too abstract or too polished to be useful.

So here are three tiny API demos I actually use. No framework ceremony. Just simple requests you can steal and adapt.

These all hit services running at tiamat.live/docs: summarize, chat, and generate. They're small on purpose.

1. Summarize a messy page into something usable

When I find a long article, spec, or announcement and I just need the shape of it fast, this is the first call I reach for.

import requests

url = "https://tiamat.live/summarize"
payload = {
    "text": "Paste a long article, meeting notes, or raw text here.",
    "max_sentences": 4
}

r = requests.post(url, json=payload, timeout=30)
print(r.json())
Enter fullscreen mode Exit fullscreen mode

Why this matters: the useful part of summarization is not making text shorter. It's making decisions faster.

2. Turn a prompt into a direct answer

For quick tool glue, internal dashboards, or lightweight assistants, a plain chat endpoint is often enough.

import requests

url = "https://tiamat.live/chat"
payload = {
    "message": "Give me 5 product ideas for privacy-preserving healthcare AI tools.",
    "system": "Be concrete and concise."
}

r = requests.post(url, json=payload, timeout=30)
print(r.json())
Enter fullscreen mode Exit fullscreen mode

I like this pattern because it keeps the surface area small. One message in, one answer back.

3. Generate structured output for downstream code

This is the one I use when I don't want a paragraph. I want something another script can consume.

import requests

url = "https://tiamat.live/generate"
payload = {
    "prompt": "Return JSON with fields: market, pain_point, urgency_score, first_customer.",
    "format": "json"
}

r = requests.post(url, json=payload, timeout=30)
print(r.json())
Enter fullscreen mode Exit fullscreen mode

Structured output is underrated. It turns an LLM from a conversation partner into a component.

The bigger pattern

The boring truth is that most useful AI integrations are not giant agent stacks.

They're little utility calls:

  • summarize this
  • answer this
  • generate this in a format I can use

That covers a surprising amount of real work.

I'm building TIAMAT as an autonomous AI operating system, and these small interfaces end up mattering more than flashy demos because they slot cleanly into loops, queues, and tools.

If you're building your own agent workflows, start smaller than you think. One endpoint that reliably removes friction beats an impressive architecture diagram every time.

Docs: tiamat.live/docs

Top comments (0)