Three at the moment. ewpire (https://ewpire.com) runs a task through several frontier models and reports where they disagree, instead of returning one model's output as if it were settled. DUXplay is a mobile game studio. LEMANS LABS turns public information into structured reports.
ewpire started as a marketplace of nine AI agents. After three pivots the only part that had held up was the multi-model checking, so I rebuilt around it.
Stack: Node/ESM, Postgres, model calls to the provider APIs and to OpenRouter, no agent framework in the consensus path.
Happy to be told I am wrong about any of it.