Plain Python Agent

import requests import anthropic client = anthropic.Anthropic() def search(query: str, platform: str = "google") -> dict: return requests.post( "https://api.scavio.dev/api/v1/search", headers={"x-api-key": "YOUR_KEY"}, json={"query": query, "platform": platform}, ).json() tools = [{"name": "search", "description": "Search the web", "input_schema": {"type": "object", "properties": {"query": {"type": "string"}, "platform": {"type": "string"}}}}]

Plain Python Agent is relevant to Google, Amazon, YouTube, TikTok, Reddit. Scavio provides a unified API to access data from all of these platforms.

The plain Python agent pattern emerged as a counter-movement to framework complexity in 2025-2026. As LangChain, CrewAI, and AutoGen grew in abstraction layers, developers building production agents increasingly chose to write direct Python code instead. Architecture: (1) LLM client -- direct HTTP calls to Anthropic or OpenAI APIs using the requests library or thin official SDKs. (2) Tool definitions -- Python functions that call external APIs and return structured results. (3) Dispatch loop -- a while loop that sends messages to the LLM, checks for tool calls in the response, executes them, and appends results. (4) State management -- a list of messages (conversation history) passed to each LLM call. The entire agent runtime is typically 50-150 lines of Python. Compare to LangChain agents which import 50+ packages and use multiple abstraction layers (chains, tools, agents, memory, callbacks). Advantages in production: (1) Debugging -- when a tool call fails, the stack trace shows your code, not framework internals. (2) Dependencies -- pip install requests anthropic (2 packages) vs pip install langchain langchain-community langchain-core (50+ transitive packages). (3) Performance -- no serialization overhead, no chain execution overhead. Direct API calls are 10-30ms faster per tool call. (4) Upgradability -- when Anthropic or OpenAI change their API, you update one function. Framework users wait for the framework to update. Tool integration example with Scavio: a Python function that takes a query string, calls api.scavio.dev/api/v1/search with requests.post, and returns the JSON response. Total code: 8 lines. No wrapper classes, no decorators, no schema generation.