#Changing /reasoning in a thread changes the reasoning globally instead of that thread/session only

1 messages · Page 1 of 1 (latest)

thorny swan
#

I would suggest implementing per-thread settings for thinking effort. Referring to this reply by Teknium, https://x.com/Teknium/status/2047813157679399193?s=20 "/model <name> in the gateway is session-scoped by default and keyed on the session key (platform:chat_type:chat_id), so a switch in one Discord channel/thread doesn't affect any other channel/thread. Stored in _session_model_overrides[session_key] in gateway/run.py, separate from config.yaml."

Shouldn't the /reasoning setting follow the same per-session logic as the /model setting?

Practical scenario: I want to use the GPT 5.5 but thinking levels should vary between threads. Example: Maintenance threads = medium reasoning, production threads = high or xhigh

Currently changing /reasoning in one thread changes thinking levels globally

Teknium 🪽 (@Teknium)

@5low_motion That is already how it currently operates

from hermes:

Good news — that's already how it works. /model in the gateway is session-scoped by default and keyed on the session key (platform:chat_type:chat_id), so a switch in one Discord channel/thread doesn't affect any

X ·
fallen oasis
#

Yep, this report is correct.

Right now /model is session-scoped in the gateway, but /reasoning <level> is still writing the global agent.reasoning_effort config value. So changing reasoning in one Discord thread can affect other gateway sessions.

There’s already a narrow PR open for this here:

https://github.com/NousResearch/hermes-agent/pull/15533

That changes gateway /reasoning <level> to be session-scoped by default, adds /reasoning <level> --global for explicitly changing the global default, and adds /reasoning reset to clear the current session override.

cc@dire isle or @inland badge for a PR review.