Hermes Agent improves workflow, but local models can bottleneck it
In firsthand use, Hermes Agent handled task delegation, conversation compression, and session management more cleanly than OpenClaw. The setup ran on an M4 Pro with 24GB of memory using oMLX, while a UGREEN DXP4800 Pro device ran Docker with separate containers for Hermes Agent and SearXNG.
Telegram became the main client, with WhatsApp used earlier. About 20 quantized local model setups were tested, including Qwen3.5-9B, Gemma 4 26B-A4B, Gemma 4 12B, gpt-oss-20b, and .
For one request at a time, Gemma 4 26B was faster than Qwen 9B, at about 64 versus 38. Qwen only pulled ahead when batching multiple requests, so the main limit was not Hermes itself but the model behind it.
Key points
- Moving from OpenClaw to Hermes Agent improved delegation, compression, and session handling.
- Hermes Agent and SearXNG were run as separate .
- Telegram was the main client, while WhatsApp was used early on.
- oMLX on an M4 Pro tested many quantized local model variants.
- Gemma 4 26B was faster for single requests, but Qwen did better under batching.