@Teknium@NousResearch 5. I know you pointed to an alternative before, but adopting tracing view as analytics, in the way kimi vis does it https://t.co/zkRT4HtmLu
@petergyang Last head to head round was openclaw vs Hermes, via Sonnet, gpt 5.4, nemotron, Kimi 2.5 and hunter alpha (now mimo).
OpenClaw took 5x more time on average and 2x the tokens vs Hermes.
The third time I did the test with a different model, Hermes created the skill and the memory
@petergyang Same anecdote I mentioned before. I have kind of a personal benchmark based on Google Ads Campaigns setup. When jumping (early March) into the "claws" I tested openclaw, nano, pico, null and Hermes. For different reasons I stayed with nano, open and Hermes.
@Teknium@turbo_xo_ yo, left this on the AMA, in case it's interesting.
Really like the OpenRouter Broadcast https://t.co/Nj1FK9sSbG
And Kimi CLI Vis
https://t.co/vhJtvVZV95
For Posthog LLM analytics, a broadcast like feature, combined with wired tracing would be amazing. Maybe is already in hermes