This dataset powers our v2 version so not published yet, the v1 with no strong facial expression is in many places ..check myned.ai avatar, or nyxclaw.ai
The bottleneck isn't managing the 'army', it's the interaction latency of a single agent. An AI 'employee' that makes a customer wait 800ms for a voice response feels broken. Getting that full STT->LLM->TTS->render loop under 250ms is the real barrier. This is the problem we're focused on with Nyx: giving the agent a real-time face and voice. An agent that can't hold a conversation isn't much of an employee. Any feedback appreciated: NyxClaw.ai
This is exactly the right direction. A native desktop client is what turns these models from a tech demo into an actual tool. We've been focused on the other side of that problem: giving local models a face and a voice. We open-sourced our work as NyxClaw.ai, it will be a fun component to pair with this.
Awesome project. The Codespaces setup is a great way to lower the barrier to entry for Hermes Agent. We've felt that setup friction ourselves, and this is a smart solution.
I saw the docs mention moving to a VPS for production. Have you seen many users make that jump yet?
The approval flow is the key feature here. Most agent frameworks focus on the agent's execution loop but forget the human-in-the-loop part, which is critical for anything with real side-effects (like DevOps).