Chat your way to a custom evolutionary algorithm

What it does LLM4AD_Next is a platform for automated algorithm design that pairs large language models with evolutionary optimization. An interactive conversational agent interviews you about your problem, then generates a complete runnable pipeline—evaluator, algorithm skeleton, configuration, and debugger—so you can evolve code for your specific task without hand-writing boilerplate.

The interesting bit The platform can scan an existing repository and recommend exactly which code blocks are most promising to evolve. It also distills algorithm design patterns like EoH, ReEvo, and MOEA/D into reusable modular skills, letting autonomous coding agents assemble the evolutionary strategy from a prompt and a file path.

Key highlights

• An AI consultant bootstraps the entire evaluator-and-search pipeline from a plain-English description via an interactive chat interface.
• The AlphaEvolve Mathematics Benchmark Suite ships with 11 independently runnable mathematical optimization cases, complete with evaluators and evolved artifacts.
• Seven search methods—including Island GA, Diverse Island GA, EoH, ReEvo, MEoH, DyCA, and MCTS-AHD—are currently available.
• Modular algorithm design skills allow coding agents to autonomously run generations by reading a skill definition and a task package.
• Optional long-term memory via MindMemOS offers global, project, and task-level scopes with configurable chat and embedding model bindings.

Caveats

• Many search methods and skills are still pending migration; the current library covers roughly half of the advertised roster.
• The README is unclear on whether the LLM consultant and memory features require external API keys or support fully local model deployment.

Verdict Worth a look if you are exploring automated heuristic design or AlphaEvolve-style mathematical optimization and would rather chat than wire up evaluators by hand. Skip it if you need a mature, fully-featured evolutionary framework today—several major methods are still marked pending.