Agent Basics: from curl to tool calls

Agent Basics is a new video series about how AI agents actually work. The first episode skips the frameworks entirely and builds up from a single HTTP request to a model calling a tool that Python executes.

Watch the first episode on YouTube.

Run a model locally

Everything in the episode runs against a model on my own machine, served by LM Studio with an OpenAI compatible API.

lms server start
lms load qwen3.8-27b --context-length 32768

A simple request

The model is behind an HTTP endpoint. You send a list of messages, a system prompt and a user prompt, and you get a message back.

curl -s http://localhost:1234/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.8-27b",
    "messages": [
      {"role": "system", "content": "You answer only in rhymes."},
      {"role": "user", "content": "What is your favorite color?"}
    ]
  }' | jq '.choices[0].message'

Structured output

Passing a JSON schema in response_format constrains the model to return an object your code can parse.

"response_format": {
  "type": "json_schema",
  "json_schema": {
    "name": "agent_anatomy",
    "strict": true,
    "schema": { "type": "object", "properties": { "model": { "type": "string" } } }
  }
}

Tool calling

Describe a function in tools and the model can choose to call it. The model returns the function name and its arguments as JSON. It does not run anything.

"tools": [{
  "type": "function",
  "function": {
    "name": "get_weather",
    "description": "Get the current temperature for a city",
    "parameters": {
      "type": "object",
      "properties": { "city": { "type": "string" } },
      "required": ["city"]
    }
  }
}]

Executing tools in Python

Read the tool call, run the function, append the result as a tool message and ask the model again.

assistant = complete()
messages.append(assistant)

call = assistant["tool_calls"][0]
arguments = json.loads(call["function"]["arguments"])
result = get_weather(arguments["city"])

messages.append({"role": "tool", "tool_call_id": call["id"], "content": json.dumps(result)})
print(complete()["content"])

That is the core of every agent. The model requests, your code executes, and the result goes back into the context. Put that in a loop and you have an agent.