How to optimize the whole code repository?

[xwqsl]

Currently, openevolve only support optimize a region in a single file, for some complex algorithm or system, the code is located in the whole repository (e.g., some algorithm are implemented as a library with various variants, and the implementation logic is complex in a repository), is there any plan to support optimizing the whole code repository?

[codelion]

Today, OpenEvolve works at the single-file level with EVOLVE-BLOCK markers. For repository-level optimization, there are a couple of practical workarounds you can use right now:

Aggregation approach - Consolidate the relevant code paths into a single file for evolution, then translate the improvements back to your codebase. This is actually what researchers did in the FORMULACODE paper to use OpenEvolve on real-world projects like Astropy.

Evaluator-driven approach - Keep your repository structure intact but point your evaluator at the full repo's test suite / benchmark. Mark the specific hotspot file for evolution with EVOLVE-BLOCK markers, and let the evaluator validate against the whole codebase. This works well when you know which file contains the performance-critical code, even if the logic touches multiple modules.

Staged evolution - Evolve different files sequentially, using the output of one evolution run as input for the next. You can target different bottlenecks across files in separate passes.

Full multi-file evolution with coordinated changes across files is on the roadmap. The main design challenges are around maintaining consistency across files and managing the search space (which grows significantly with more evolve blocks). PRs and design proposals are welcome if you'd like to contribute!