How do I change the model?

[candid mural]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[tepid lava]

Based on the docs, here's what's happening and how to fix it:

The Issue

When a model runs out of quota mid-task, Clawdbot can't retry with a different model until that session's context is cleared. The stuck session is holding onto the failed request.

How to Fix

Option 1: Start a fresh conversation (quickest)

/new

This creates a new session with fresh context. The old stuck request is abandoned.

Option 2: Cancel the running task

/cancel

If the command is still processing, this aborts it and lets you retry with the new model.

Option 3: Set the new model system-wide

clawdbot models set google/gemini-2.5-flash-lite
clawdbot gateway restart

Or for on-the-fly switching in chat:

/model google/gemini-2.5-flash-lite

Why Your Config Change Didn't Work

Changing config requires a gateway restart for it to take effect. The stuck session is also independent of config — it has its own context.

Recommended Fix Order

1. Run /new in chat → fresh session
2. Run /model google/gemini-2.5-flash-lite to switch models
3. Retry your command

If the gateway is truly stuck (not responding), run:

clawdbot gateway restart

Gemini 2.5 Flash Lite should be more quota-efficient than 3 Flash Preview, so you shouldn't hit this as often once switched over.

[polar lynx]

/model google/gemini-2.5-flash-lite returns:
"Model "google/gemini-2.5-flash-lite" is not allowed. ..."

@tepid lava Can you help?

[tepid lava]

I'll help! 🦐

The issue is likely a stuck session from when the quota ran out. Try these steps in order:

1. In chat, run:

/new

This kills the stuck session and starts fresh.

2. Then switch to the new model:

/model google/gemini-2.5-flash-lite

3. Retry your command

---

If the bot isn't responding at all:

clawdbot gateway restart

Then try again in a fresh chat.

Let me know if that gets you unstuck!

[polar lynx]

"clawdbot gateway restart" does not work.
I get:
"Agent failed before reply:
All models failed (2): google/gemini-3-pro-preview: No available auth profile for google (all in cooldown or unavailable). (rate_limit)
google/gemini-3-flash-preview: No available auth profile for google (all in cooldown or unavailable). (rate_limit)"

[sleek hedge]

you are using gemini CLI auth not API right?

[outer plover]

gateway health connect failed. What to do?

connect ECONNREFUSED 127.0.0.1:18789

can someone help?

[sleek hedge]

open separate problem

[tepid lava]

Great question! There are two ways to use Gemini:

1. API Key (Standard)

• Provider: google
• Auth: GEMINI_API_KEY env var
• This is what most people use

2. Gemini CLI OAuth (What you asked about)

• Provider: google-gemini-cli
• Requires the bundled plugin:

  clawdbot plugins enable google-gemini-cli-auth
  clawdbot models auth login --provider google-gemini-cli --set-default
  

---

Which one are you using? If you set it up through the onboarding wizard (clawdbot onboard), it likely used the API key method unless you specifically enabled the Gemini CLI plugin.

You can check your auth setup with:

clawdbot models status

[polar lynx]

I am using API

[tepid lava]

The gateway isn't running. Start it with:

clawdbot gateway start

Or if you want it to run in the background as a service:

clawdbot gateway start && clawdbot gateway status

Then try connecting again. The ECONNREFUSED error just means nothing is listening on port 18789.

Correct — Muneeb216's gateway issue is a different problem. Please start a new thread in #help for that!

For reference, quick fixes to try:

• clawdbot gateway start
• clawdbot status
• Check if another process is using port 18789: lsof -i :18789

---

Back to the original issue — Bit-EXE, let me know which Gemini auth method you're using (API key vs CLI) and we can sort out the model switch!

[sleek hedge]

ah, not sure then, if you want to restart local gateway, clawdbot daemon restart

[mossy nova]

Can I use Grok to run clawd?

[tepid lava]

Perfect — so you're using google with GEMINI_API_KEY. Here's the fix:

Quick Steps

1. Clear the stuck session (run in chat):

/new

2. Switch to the new model:

/model google/gemini-2.5-flash-lite

3. Retry your command

---

If That Doesn't Work

Check your config and ensure the model is allowed:

clawdbot models status

You might need to add it to your allowlist if you have agents.defaults.models set. The model reference should be:

• google/gemini-2.5-flash-lite

Gemini 2.5 Flash Lite is much more quota-friendly, so you shouldn't hit this issue as often! Let me know if it works. 🦐

[mossy nova]

Can I use Grok to run clawd? @tepid lava

[tepid lava]

Flip3's right — if /new doesn't unstuck it, try restarting the gateway:

clawdbot gateway restart

Then start a fresh chat and try again.

Summary of steps to try in order:

1. /new in chat
2. If still stuck → clawdbot gateway restart
3. /model google/gemini-2.5-flash-lite
4. Retry your command

Let me know what happens! 🦐

[sleek hedge]

just to make sure, you also configured the model via clawdbot onboard? besides selecting it, you need to configure first

[tepid lava]

Yes! xAI Grok is supported.

• Provider: xai
• Auth: XAI_API_KEY
• Model example: xai/grok-3

Set it up via:

clawdbot onboard --auth-choice xai-api-key

Or manually:

export XAI_API_KEY="your-key-here"
clawdbot models set xai/grok-3

Check available Grok models with:

clawdbot models list --provider xai

[polar lynx]

How do I do that?

[tepid lava]

Good catch by flip3! Make sure you actually configured Gemini via the wizard, not just selected it.

Run through the onboarding:

clawdbot onboard

Or specifically configure the model:

clawdbot models set google/gemini-2.5-flash-lite

Then check:

clawdbot models status

Does it show Google/Gemini with valid auth? If not, that's why it's stuck.

[sleek hedge]

krill is right here, go through onboard again, just select to only update values, and select the model you want to config

[tepid lava]

Exactly what flip3 said! 🦐

Run:

clawdbot onboard

It will ask if you want to update existing config or start fresh — choose update existing.

Then:

1. Select Google/Gemini as the provider
2. Enter your API key
3. Pick Gemini 2.5 Flash Lite

That's it! The wizard handles all the config for you.

Let me know if you hit any snags!

[polar lynx]

OK. I managed to change my Gemini 3 Pro to the Gemini 2.5 Flash Lite, but it still says I am out of quota. Guess I will have to wait for it to reset and in the future stick to 2.5 Flash Lite.

[sleek hedge]

cheers, you can also check how much you used, but i am not sure why you would run into quota issues with an API key? dont you just pay as you use?

[polar lynx]

It is a free key so it has limits. But I am not sure why I am getting them still since I changed to a new API key from a different Google account. I initially started with an API key from my free account on Google, but later wanted to switch over to my Pro paid account with higher quotas. I gave it a new API key from my paid account, but it still oddly enough says Ive hit my quota limits. Both keys are free, not billed, so they have limits, but I think the Pro account has higher limits even on free API keys.

Edit: I am wrong, they Pro free key has the same limits. Still odd that it doesnt reset.

[tepid lava]

That's correct — quota is tied to your Google Cloud billing account, not the specific model. So even switching models won't help once you've hit the limit.

The good news: Gemini 2.5 Flash Lite is much more efficient, so you should hit the limit less often going forward. It's designed for high-volume, low-cost usage.

To check your quota:

• Go to Google Cloud Console → Vertex AI → Quotas
• Look at "Request counts" for the Gemini API

Tips to avoid hitting it:

• Use 2.5 Flash Lite going forward (you've already switched!)
• Set up fallback models so you don't get completely stuck

Feel free to ask in #models if you want suggestions for a cheap fallback model to pair with Gemini! 🦐

[tepid lava]

Flip3 raises a good point! With a standard API key, it's usually rate limits (RPM - requests per minute), not usage limits.

Google's free tier has:

• Rate limits on requests per minute
• Token limits per minute

Once you enable billing, you get higher limits — but you can also set spending caps in Google Cloud Console if you want to cap usage.

Gemini 2.5 Flash Lite being more efficient helps because:

• Fewer tokens per request = less throughput
• Faster response times = fewer concurrent requests

So it's likely a rate limit issue, not a cost issue. Enabling billing on your Google Cloud account should raise those limits significantly.

[sleek hedge]

use the sub of the pro account with antigravity/CLI auth tokens then, i dont know the free tier, but it gets me quite far

skip the API option and select the google CLI option and/or antigravity in the next step

[polar lynx]

I am waiting 10 minutes for the timeout to finish and see if I can get this API key version working first. I changed it over to a paid key but it is still stuck.

[sleek hedge]

try using /stop

[polar lynx]

/ stop and / new still hit the limit

[tepid lava]

That's odd! Two things to check:

1. Verify the new API key is actually being used:

clawdbot models status

Does it show your paid account? If not, the old key might still be cached.

2. Check where your API key is stored:

clawdbot config get | grep -A5 google

3. Make sure billing is enabled on the new project:

• Go to Google Cloud Console → Your Project → Billing is linked
• Also check that Vertex AI API is enabled in the project

[polar lynx]

It's out of timeout and using 2.5 gemini flash with my paid key now, hopefully no more timeouts

[tepid lava]

The most common issue: creating a new API key doesn't automatically link it to your paid billing account. You need to:

1. Create the key in the same Google Cloud project that has billing enabled
2. Or enable Gemini API in the new project

Try creating a new API key in your paid Google Cloud project and update it:

clawdbot onboard

(Select Google again and paste the new key)

Let me know what clawdbot models status shows!

[tepid lava]

That's not normal — /new should give you a fresh session. Something's still wrong with the config.

Let's debug this:

1. Check which API key is loaded:

clawdbot models status

2. Restart the gateway to clear any cached auth:

clawdbot gateway restart

3. Verify your config has the right key:

[tepid lava]

Flip3 has a better solution! If you have a Google Pro subscription, you can use the Antigravity OAuth or Gemini CLI auth which comes with your subscription and has much higher limits than free-tier API keys.

Two Options:

Option A: Gemini CLI Auth (Recommended)

clawdbot plugins enable google-gemini-cli-auth
clawdbot models auth login --provider google-gemini-cli --set-default

Option B: Antigravity OAuth

clawdbot plugins enable google-antigravity-auth
clawdbot models auth login --provider google-antigravity --set-default

Both of these use your Google Pro subscription credentials instead of a free-tier API key, so you get the higher quotas that come with Pro.

[tepid lava]

Glad it's working now! 🎉

Gemini 2.5 Flash Lite should be much more stable:

• More efficient = less quota usage
• Faster responses = fewer timeouts
• Paid key = higher limits

If you still hit issues down the road, remember the Antigravity/CLI OAuth options flip3 mentioned — they come with Pro subscription benefits and higher quotas.

Happy chatting! 🦐

clawdbot config get | grep -A10 "google"

4. If needed, manually set the env var:

export GEMINI_API_KEY="your-new-paid-key"
clawdbot gateway restart

The most likely issue: your GEMINI_API_KEY env var is still pointing to the old free account key, not the new paid one.

Check echo $GEMINI_API_KEY — does it show the right key?