pop-framework 1.2.0

Fast, lean AI agents. 5 lines to production.

Coding agents: See SKILLS.md for the complete API guide — tools, agents, multi-agent patterns, streaming, memory, and all imports in one file.

pop is a lightweight Python framework for building AI agents. It supports multiple LLM providers, has 5 core concepts, and gets you from install to a working agent in under 2 minutes.

Why pop?

• 5 lines to a working agent -- define a tool, create an agent, call run.
• 8 LLM providers built-in -- OpenAI, Anthropic, Gemini, DeepSeek, Grok, Kimi, MiniMax, GLM. Switch by changing one string.
• ~2,500 lines of code -- read the entire framework in an afternoon.
• 2 runtime dependencies -- httpx and pydantic. Import time under 1ms (lazy imports).
• Zero commercial dependencies -- no forced telemetry, no vendor lock-in.

Install

uv add pop-framework
# or
pip install pop-framework

All 8 providers (OpenAI, Anthropic, Gemini, DeepSeek, Grok, Kimi, MiniMax, GLM) are included — no extras needed.

Quick Start

from pop import Agent, tool

@tool
def search(query: str) -> str:
    """Search the web for current information."""
    return web_search(query)  # your implementation

agent = Agent(model="openai:gpt-4o", tools=[search])
result = agent.run("What happened in AI today?")
print(result.output)

That's it. No StateGraph, no RunnableSequence, no ChannelWrite.

Docs

Benchmarks

Reproduce: python benchmarks/bench_startup.py && python benchmarks/bench_dx.py && python benchmarks/generate_charts.py

Details: docs/benchmarks.md

License

MIT