AgentOpt v0.1.0
Find the right LLM models for your AI agents — automatic model selection with accuracy/cost/latency tradeoffs.
Highlights
- Duck-typed agent API: wrap your agent with
__init__(self, models)+run(self, input_data)— no base class required - 9 selection algorithms: brute force, arm elimination, epsilon-LUCB, hill climbing, Bayesian optimization, and more
- Framework-agnostic: works with OpenAI, LangChain, LangGraph, CrewAI, LlamaIndex, AG2, and any httpx-based SDK
- Automatic cost & latency tracking: intercepts at the httpx transport layer
- Response caching: in-memory + SQLite on disk, enabled by default
- Parallel evaluation: concurrent agent runs with configurable concurrency
Install
pip install agentopt