light sluice
Switching over from Openclaw and using Requesty as the API provider. Started a new install and entered all the information in the setup and it gives me this error when I try to say hi in the TUI. This is my model setup code snippet. Any help greatly appreciated.
model:
default: vertex/gemini-2.5-flash-lite@us-west1
provider: custom
base_url: https://router.requesty.ai/v1
api_key: rqsty-sk-XXXXX
fallback_providers: []
meager knoll
This is probably not your install.
The error means Requesty's Gemini route is rejecting Hermes' normal tool list. Hermes exposes many tools up front, and that backend appears to only allow multiple tools in very limited cases.
Quick test:
hermes chat --toolsets search -q "hi"
That starts Hermes with only the search toolset (web_search only).
You can also try:
hermes chat --toolsets memory -q "hi"
If one of those works, but the normal default setup does not, that points to the provider/model route being the issue rather than your config syntax.
Most likely fix:
Use a different Requesty route that has better OpenAI-style function/tool calling support, such as an OpenAI or Claude model route.
So the short version is:
vertex/gemini-2.5-flash-lite@us-west1 with Hermes' normal tool setuphermes chat --toolsets search -q "hi" is the quickest confirmation
light sluice
THANK YOU for such a detailed reply. Appreciate it. You are right, running both those commands you suggested works. Switching to something like Deepseek v3.2 does not cause any issues.
Are you able to explain why the cheaper gemini flash model does not work with Hermes? I was using this on OpenClaw fine so I thought it would be useable on Hermes.
Also, what is the suggestion for summary_model and summary_provider? Should they also match the default model or can it be something else cheaper and smaller?
meager knoll
it's probably just not equipped to do tool calling or not tagged as capable from the provider. the newer flash models from google are pretty good at that kind of thing for cheap. they can def be something cheaper than your main model.
light sluice
@meager knoll should I also be using the same model for summary_model?