wtf is wolbarg-benchmarks?

Wolbarg Benchmarks is a companion repository that publishes public, reproducible performance numbers for Wolbarg, a separate software development kit that gives AI agents local-first semantic memory. Semantic memory here means a system that lets an AI agent store and later recall information based on meaning rather than exact keyword matches, running on the user's own machine rather than a remote server. This repository does not contain the memory SDK itself, it only measures how fast that SDK performs under different conditions. The benchmarks test the complete storage and recall path of the SDK, backed by either SQLite or PostgreSQL, and report numbers such as how long it takes the system to become ready from a cold start, how many memory writes per second it can sustain, and how quickly it can recall relevant memories once a couple thousand items have been stored. The README is careful to note that these numbers use mock embeddings rather than a live AI provider, so results will vary by machine and should not be mixed up with a separate suite that tests real provider-generated embeddings. Anyone curious about Wolbarg's real-world performance, or comparing it against other agent memory systems, can clone this repository, install its dependencies, and run the benchmark script themselves to reproduce the same measurements on their own hardware. It also includes a feature comparison file that lays out which agent-memory capabilities the current version of Wolbarg supports, such as write-time deduplication, real-time change events, and multi-writer concurrency handling. The repository is released under the MIT license.