#Tool does not exist

1 messages · Page 1 of 1 (latest)

pulsar juniper
#

Just got hermes setup iin docker. I'm using a local Ollama instance, also in docker. I'm running gemma4:31b-it-q4_K_M Haven't been able to do much yet, and don't know enough to solve it. Is this a problem of the model?

image.png image.png
slender jungle
#

That first screenshot is the model trying to call a tool that Hermes does not have: messaging:list_channels.

Your second screenshot is a good sign though: Hermes received the Discord message and replied with working, so Docker/Discord connectivity itself is working.

This is most likely the local Ollama model inventing a tool name instead of using one of the exact tools Hermes gave it. Some local models can chat normally but still be unreliable with tool calls.

Try a simpler tool-call test that does not depend on Discord channel permissions:

Use the terminal tool to run: pwd

If that works, Hermes tool calling is generally working and this was the model making up a messaging tool name. If it fails or invents another tool name, try a stronger tool-calling model for Hermes, or temporarily switch models to confirm.

For Discord message targets specifically, Hermes uses the send_message tool. Channel/server admin listing is a separate Discord admin capability and needs the right toolset and bot permissions.

pulsar juniper
#

Even that simple cmd doesn't work.

I'll try a new model. I've pulled one locally into ollama, how can I have hermes refresh the list of available models that ollama has?

image.png
slender jungle
#

We need logs/details before we can tell whether this is the model, Ollama’s tool-call support, the configured API mode, or something Docker-specific.

The first screenshot showed the model inventing messaging:list_channels, which is not a Hermes tool. The later pwd test is useful, but the bot replying conversationally instead of calling terminal still points at a model/tool-calling path issue. We need the actual runtime details to troubleshoot that.

Please post a fresh /debug from the same setup, plus:

hermes --version
hermes model
hermes dump

Also include the exact Ollama model name you switched to, how Hermes is pointed at Ollama in Docker, and the relevant agent.log lines from the failed Use the terminal tool to run: pwd attempt.

For the Ollama model list question: after pulling a model into Ollama, restart the Hermes process/container or re-open the model picker so Hermes re-queries the Ollama endpoint. If it still does not show up, include the output of ollama list from the Ollama container/host and the Ollama base URL Hermes is using.