Invalid API key.. Fix external API key

[gigachad80]

Openclaude Version

0.1.4

.env which I set

export CLAUDE_CODE_USE_GEMINI=1
export GEMINI_API_KEY=AIxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
export GEMINI_MODEL=gemini-3.1-flash-lite-preview
export ANTHROPIC_API_KEY=dummy

Note

I'm not using Openrouter / OpenAI Base URL

The commands which I used :

export CLAUDE_CODE_USE_GEMINI=1
export GEMINI_API_KEY=AIxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
export GEMINI_MODEL=gemini-3.1-flash-lite-preview
export ANTHROPIC_API_KEY=dummy
openclaude

Note

I also didn't give any spaces or used any double quotes

Output : #43 (comment)

Invalid API key.. Fix external API key

Here's the proof that the API key is valid : #43 (comment)

[Vasanthdev2004]

Thanks for the report. I don't think this necessarily means your Gemini key is bad.

A couple things stand out:

• you're on 0.1.4, and the repo is now on 0.1.6
• Invalid API key ? Fix external API key is a generic external-provider auth message in OpenClaude, so on older versions it can be misleading
• gemini-3.1-flash-lite-preview also doesn't look like a current documented Gemini OpenAI-compatible model ID

Could you retry on the latest version first, with a known supported Gemini model?

npm install -g @gitlawb/openclaude@latest

export CLAUDE_CODE_USE_GEMINI=1
export GEMINI_API_KEY=YOUR_REAL_KEY
export GEMINI_MODEL=gemini-2.5-flash
unset ANTHROPIC_API_KEY
openclaude

Google's current OpenAI-compat docs use gemini-2.5-flash with the Gemini OpenAI endpoint:

• https://ai.google.dev/gemini-api/docs/openai
• https://ai.google.dev/gemini-api/docs/models/gemini-v2

If it still fails on 0.1.6 with gemini-2.5-flash, please paste the new full error output. That would make this look like a real current Gemini auth/compatibility bug rather than an old-version or model-ID issue.

[gigachad80]

Hey @Vasanthdev2004 , as I said I'm not using OpenAI or Openrouter . I'm just using Gemini API and model
After running those command which you suggested , it's asking me to login and there's no login command
For more info , check this : #43 . You'll get to know that why I'm using without OpenAI and why there's no login command

[davccavalcante]

Hey @Vasanthdev2004 , as I said I'm not using OpenAI or Openrouter . I'm just using Gemini API and model After running those command which you suggested , it's asking me to login and there's no login command For more info , check this : #43 . You'll get to know that why I'm using without OpenAI and why there's no login command

1. Obtain your API key from https://openrouter.ai/workspaces/default/keys. Create an account if you do not have one, and bla, bla, bla...

2. Install the OpenCLAUDE tool globally in your terminal:

npm install -g @gitlawb/openclaude@latest

3. Set the following environment variables in your terminal:

   Option 1: Using OpenAI compatible endpoints

   export CLAUDE_CODE_USE_OPENAI=1
   export OPENAI_API_KEY='YOUR_KEY_HERE'
   export OPENAI_BASE_URL='https://openrouter.ai/api/v1'
   export OPENAI_MODEL='stepfun/step-3.5-flash:free'

   Option 2: Using Gemini compatible endpoints

   export GEMINI_MODEL='qwen/qwen3.6-plus-preview:free'

4. Run the OpenCLAUDE command:

openclaude

David C Cavalcante
https://x.com/davccavalcante
https://linkedin.com/in/hellodav
https://github.com/davccavalcante

[salmanrajz]

This appears to be fixed in v0.1.6. The current isAnthropicAuthEnabled() in src/utils/auth.ts correctly checks for CLAUDE_CODE_USE_GEMINI and disables Anthropic auth when a 3P provider is active. The useApiKeyVerification hook returns 'valid' immediately in that case, so the 'Invalid API key' error shouldn't appear.

Upgrading to the latest version should resolve this:

npm install -g @gitlawb/openclaude@latest

[Vasanthdev2004]

You're right ? I wasn't telling you to use OpenAI or OpenRouter as your provider.

What I meant is: OpenClaude's Gemini support is implemented through its OpenAI-compatible shim internally, but the user-facing setup is still Gemini:

• CLAUDE_CODE_USE_GEMINI=1
• GEMINI_API_KEY=...
• GEMINI_MODEL=...

So if openclaude is still sending you to login, that's not expected for a working Gemini setup.

Given #43, the most likely explanation is that you're still hitting an older/stale install rather than the current 0.1.6 code path.

Could you please check these exact commands and paste the output?

openclaude --version
which openclaude
npm ls -g @gitlawb/openclaude --depth=0

If you want to bypass any stale global install completely, try this once:

CLAUDE_CODE_USE_GEMINI=1 \
GEMINI_API_KEY=YOUR_REAL_KEY \
GEMINI_MODEL=gemini-3.1-flash-lite-preview \
npx @gitlawb/openclaude@latest

If that still drops you into login, then this is a real current Gemini startup regression and not just an old install/version mismatch.

[gigachad80]

Answering your questions , here's the output

• 0.1.6 (Open Claude)

• /c/Users/Dark/AppData/Local/pnpm/openclaude
  Legend: production dependency, optional only, dev only

• C:\Users\Dark\AppData\Local\pnpm\global\5
  dependencies:
  @gitlawb/openclaude 0.1.6

Also using npx , it still asking me to login

[salmanrajz]

Dug into this further. The root cause is that setting ANTHROPIC_API_KEY=dummy triggers the Anthropic auth validation path even when using Gemini. The error handler in src/services/api/errors.ts sees the dummy key as an external API key source and shows 'Invalid API key'.

Workaround: Don't set ANTHROPIC_API_KEY at all when using Gemini. It's not needed — just set:

export CLAUDE_CODE_USE_GEMINI=1
export GEMINI_API_KEY=your_key_here
export GEMINI_MODEL=gemini-3.1-flash-lite-preview

The isAnthropicAuthEnabled() function correctly disables Anthropic auth when CLAUDE_CODE_USE_GEMINI is set, but having ANTHROPIC_API_KEY in the environment can still trigger validation in certain code paths. Removing it should fix the issue.

[davccavalcante]

I think the problem lies between the chair and the computer. Z)

First, test in another folder, export as I explained, use the OpenRouter key and run the command to open TUI. Also, you don't need to use commands like @latest and others; you can uninstall using npm if you installed it globally, and then reinstall it globally. Remember that each time you use the CLI/TUI in a specific folder, you will need to export again, or you can add it directly to your .zshrc file, for example, if you use one. This way it will work in any folder, and you will only have to run the command that opens TUI.

You will see like this:

1. Obtain your API key from https://openrouter.ai/workspaces/default/keys. Create an account if you do not have one, and bla, bla, bla...

2. Install the OpenCLAUDE tool globally in your terminal:

npm install -g @gitlawb/openclaude

3. Set the following environment variables in your terminal:

   Option 1: Using OpenAI compatible endpoints

   export CLAUDE_CODE_USE_OPENAI=1
   export OPENAI_API_KEY='YOUR_KEY_HERE'
   export OPENAI_BASE_URL='https://openrouter.ai/api/v1'
   export OPENAI_MODEL='stepfun/step-3.5-flash:free'

   Option 2: Using Gemini compatible endpoints

   export GEMINI_MODEL='qwen/qwen3.6-plus-preview:free'

4. Run the OpenCLAUDE command:

openclaude

David C Cavalcante
https://x.com/davccavalcante
https://linkedin.com/in/hellodav
https://github.com/davccavalcante

[gigachad80]

Hey @davccavalcante . Is there any way to use openclaude without purchasing credits. Also I'm not using Win., so it doesn't have .bashrc/.zshrc .I'm using Openclaude via terminal emulator instead of Powershell & CMD . Should I reinstall it in new directory ?

[Vasanthdev2004]

@gigachad80 Hey — you do not need OpenAI or OpenRouter for this.

And no, you do not need to reinstall in a new directory just for this. The current folder usually is not the root problem here.

What matters much more is the environment in the same shell session where you launch openclaude.

Based on the latest debugging in this thread, the most important part is:

• do not set ANTHROPIC_API_KEY at all when using Gemini

So the cleanest test would be:

env -u ANTHROPIC_API_KEY \
  CLAUDE_CODE_USE_GEMINI=1 \
  GEMINI_API_KEY='YOUR_REAL_KEY' \
  GEMINI_MODEL='gemini-2.5-flash' \
  openclaude

That avoids any stale Anthropic key leaking into the startup path.

About credits:

• if you want truly zero-cost, the most reliable option is a local model via Ollama
• if you want to stay on Gemini only, Google still has a free tier for some Gemini API usage depending on model/quota:
  Gemini pricing
  Gemini billing

So yes, it is possible to use OpenClaude without purchasing OpenAI/OpenRouter credits.

And for your shell setup:

• if your terminal emulator doesn't use .bashrc / .zshrc, that's okay
• you can just set the vars in the current session before launching
• or use the one-shot env ... openclaude command above

If that one-shot command still sends you to login, then this is still a real current Gemini startup bug and not just shell config.

[davccavalcante]

@gigachad80

Just do what I told you to do, it's simple, you don't need to pay anything for OpenRouter, I sent you the link, the settings, etc. If you're struggling with something simple, with Ollama, with LM Studio, you're going to suffer even more. You DON'T even need to use the Gemini API, just do exactly what I taught you.

Last time, dude.

1. Obtain your API key from https://openrouter.ai/workspaces/default/keys. Create an account if you do not have one, and back to https://openrouter.ai/workspaces/default/keys and create YOUR API KEY (YOUR_KEY_HERE), and COPY your API KEY, DO NOT SHARE.

bash = RUN THE COMMAND IN YOUR TERMINAL, like Warp, iTerm, etc.

2. Install in your terminal:

npm install -g @gitlawb/openclaude

3. Copy the following and Paste in your terminal:

YOU CAN CHOICE YOUR MODEL HERE FOR FREE https://openrouter.ai/models?fmt=cards&input_modalities=text&max_price=0&output_modalities=text

Option 1 WITH stepfun LLM endpoints

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=YOUR_KEY_HERE
export OPENAI_BASE_URL=https://api.groq.com/openai/v1
export OPENAI_MODEL=stepfun/step-3.5-flash:free

Option 2 WITH qwen LLM endpoints

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=YOUR_KEY_HERE
export OPENAI_BASE_URL=https://api.groq.com/openai/v1
export OPENAI_MODEL=qwen/qwen3-32b

Option 3 WITH qwen LLM endpoints I USE THIS

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=YOUR_KEY_HERE
export OPENAI_BASE_URL=https://api.groq.com/openai/v1
export OPENAI_MODEL=qwen/qwen3.6-plus-preview:free

4. Run in your terminal the command:

openclaude

You will see like this:

David C Cavalcante
https://x.com/davccavalcante
https://linkedin.com/in/hellodav
https://github.com/davccavalcante

[gigachad80]

@Vasanthdev2004 Now it's asking me to run/login but there's no login option

[gigachad80]

Lemme try @davccavalcante ur method

[davccavalcante]

[gigachad80]

Just saw that you mentioned OpenAI . Never created account on OpenAI platform and I don't want to . Claude & Gemini are way etter than GPT models. Any other way ?