AgentOptimizer

AgentOptimizer / agentopt

Public

AgentOpt automatically finds the best LLM model combination for each step of your agent — optimizing for accuracy, cost, and latency.

agentoptimizer.github.ioagentopt
18
0
100% credibility
Found Mar 23, 2026 at 18 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

AgentOpt is a Python library that automatically finds optimal combinations of large language models for multi-step AI agents by evaluating them on a user-provided dataset for accuracy, cost, and latency trade-offs.

How It Works

1
🔍 Hear about AgentOpt

You learn about a smart tool that helps pick the perfect AI brains for your helper bot to save money and time.

2
📥 Get the tool ready

You bring the tool into your project with a simple download, and it's all set up in moments.

3
🤖 Show your AI helper

You describe the steps in your AI helper, like planning and solving, so the tool knows what to optimize.

4
📝 Share test examples

You give a handful of real questions and answers from your work, like a practice set for the tool to learn from.

5
⚡ Start the search

You tell the tool to find the best mix of AI brains by running quick tests on your examples.

6
📊 Review top choices

You get a clear table ranking the best options by smarts, speed, and cost savings, like 10-100 times cheaper.

🎉 Upgrade your bot

Your AI helper now runs faster and costs way less, handling tasks perfectly without breaking the bank.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 18 to 18 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is agentopt?

AgentOpt automatically finds the best LLM model combination for each step in your agent, optimizing for accuracy, cost, and latency. Give it your agent class, a small eval dataset, and candidate models per step—like planner: gpt-4o vs gpt-4o-mini—and it searches the space efficiently, outputting a Pareto curve of tradeoffs. Built in Python, it intercepts httpx calls from any framework, tracking tokens and latency without proxies or heavy wrappers.

Why is it gaining traction?

Unlike manual tuning or single-model routers like AgentOps, AgentOpt handles multi-step agents agentically, slashing costs 20-100x on benchmarks like HotpotQA while matching accuracy. Smart methods from brute-force to Bayesian optimization prune bad combos early, with parallel evals, response caching, and exports to CSV/YAML. Devs love the quickstart: pip install, wrap your agent, run selector.select_best().

Who should use this?

Agent builders on LangChain, LangGraph, CrewAI, or LlamaIndex tuning multi-LLM pipelines for prod. Ideal for teams evaluating Claude, GPT, or Gemini combos on custom datasets, especially when latency/cost matter more than perfect scores. Skip if your agent is single-model or you need Node.js agentoptions support.

Verdict

Try it for agent optimization—solid docs, examples for major frameworks, and Apache 2.0 make it low-risk despite 18 stars and 1.0% credibility signaling early alpha. Production? Wait for more evals, but great for dev prototyping.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.