OpenClaw combines vector and keyword search using a weighted score fusion strategy. By default, vector results get 70% weight and keyword results get 30%. The candidateMultiplier setting controls how many candidates each search path retrieves before fusion.
With a candidateMultiplier of 3 and a final result limit of 10, each path fetches 30 candidates. The fusion step merges both lists, applies weights, deduplicates, and returns the top 10. You can tune the weights in your config if your workflow favors exact matches over semantic ones or vice versa.