#gemini 3 flash from google ai studio api key isnt working as an agent

1 messages · Page 1 of 1 (latest)

autumn sigil
#

I need to keep prompting it to getting doing things, it also explains everything he is thinking

crimson flickerBOT
#

What you’re describing is usually (a) “reasoning is being shown” and/or (b) the model is planning but not actually tool-calling.

1) Stop it from “explaining what it’s thinking”

In the channel/thread where you talk to the agent, send these as standalone messages:

  • /reasoning off ← hides the “Reasoning:” / thinking blocks
  • /verbose off ← stops the extra tool/debug bubbles (if you enabled them)

Docs: https://docs.openclaw.ai/tools/thinking (covers /reasoning, /verbose, /think)

If it’s still “thinking out loud” inside the normal answer text, that’s not OpenClaw’s reasoning display — that’s the Gemini model outputting it directly. In that case the fix is prompt/model-side (e.g. add “Don’t reveal internal reasoning, just give the result” to your agent/system prompt, or switch to a model that supports separated/hidden reasoning).

2) “I need to keep prompting it to do things”

That usually happens when tools are unavailable/blocked or the model is weak at tool-calling.

Quick checks:

  • Do you see errors like “Tool 
 not allowed” or “Tool 
 not found” in your gateway logs?
#
  • When you ask it to do something concrete (web fetch, browser, exec, etc.), does it actually call tools or just describe steps?

If you paste:

  1. openclaw --version
  2. what exact model/provider you’re using (the /model status output is perfect)
  3. a short snippet of a reply where it “thinks” + the relevant gateway log lines


I can tell you which of the above cases you’re in and what to change.