Technical writing from the team about routing internals, benchmark data, EU compliance, and the rare opinion piece. Quiet by design, because we'd rather publish three good things a year than thirty mediocre ones.
We ran the same 16-category eval across qwen2.5-coder, deepseek-r1, gemma3, phi4-mini and the qwen3 family on three Apple Silicon configs. The results say less about the models and more about hardware sensitivity.
Read more →Open the rate-limit table of any cloud AI provider and you find a meter on everything. Running local-first by default makes most of those limits disappear.
Read more →Most AI tools treat data protection as something you configure. With a local-first architecture, the safe state is the default one, because the data never leaves the machine to begin with.
Read more →If you can’t reproduce a routing decision, you can’t audit it. This is how the router stays deterministic without giving up score-based dispatch.
Read more →Local-first architecture solves most of the Act’s compliance asks by default. We walk through which articles map to which engine guarantees.
Read more →