Windows/Gemini: "404 v1main" on native provider & "No API Key" error when forcing OpenAI provider

[TypicalPlay]

Description:
I am trying to run picoclaw on Windows 11 using the Google Gemini API. I have encountered two blocking issues:

Native Gemini Provider: Fails with a 404 error because it targets the deprecated v1main endpoint.

OpenAI Bridge Workaround: Fails with a "no API key configured" error when I try to use the openai provider with generativelanguage.googleapis.com as the base URL, even though the key is correctly defined in config.json.

Steps to Reproduce:

Scenario A: Native Gemini Provider

Configure config.json with provider: "gemini" and model: "gemini-1.5-flash" (or gemini-pro).

Run .\picoclaw.exe gateway.

Send a message via Telegram.

Result: Error 404 "models/gemini-1.5-flash is not found for API version v1main".

Scenario B: OpenAI Provider (Workaround)

Configure config.json with provider: "openai", api_base pointing to Google's OpenAI-compatible endpoint, and model: "gemini-1.5-flash".

Run .\picoclaw.exe gateway.

Result: Immediate crash on startup: Error creating provider: no API key configured for provider (model: gemini-1.5-flash).

Configuration (config.json):
(Note: I have tried api_key, apiKey, and token keys in the JSON, but none are detected).

{
"agents": {
"defaults": {
"workspace": "./workspace",
"provider": "openai",
"model": "gemini-1.5-flash",
"max_tokens": 8192,
"temperature": 0.7
}
},
"providers": {
"openai": {
"api_key": "AIzaSy...",
"api_base": "https://generativelanguage.googleapis.com/v1beta/openai/"
}
},
"channels": {
"telegram": {
"enabled": true,
"token": "YOUR_BOT_TOKEN",
"allow_from": ["YOUR_USER_ID"]
}
},
"gateway": {
"host": "0.0.0.0",
"port": 18790
}
}

Logs / Error Output:

Scenario A (Native Gemini):

API error: [{
"error": {
"code": 404,
"message": "models/gemini-1.5-flash is not found for API version v1main, or is not supported for generateContent...",
"status": "NOT_FOUND"
}
}]
Scenario B (OpenAI Provider):

PS D:\PicoClaw> .\picoclaw-windows-amd64.exe gateway
Error creating provider: no API key configured for provider (model: gemini-1.5-flash)
Environment:

OS: Windows 11

Version: Latest Release (picoclaw-windows-amd64.exe)

Model: Google Gemini 1.5 Flash / Pro

Expected Behavior:

The native Gemini provider should use the current Google API endpoints (v1beta/v1) instead of v1main.

Alternatively, the OpenAI provider should correctly read the API key from config.json even if the model name contains "gemini", allowing us to use Google's OpenAI-compatible endpoint.

[algo-uno-10]

use gemini 2.5 flash @TypicalPlay

[TypicalPlay]

Will make the switch. Thank you.

[Pascal01dev]

Many Go-based LLM tools (like PicoClaw) use a generic provider structure. Instead of nesting the key under openai, try defining the Google OpenAI-compatible endpoint as a standalone custom provider if the tool supports it

[TypicalPlay]

Thanks, will check that option too.

[nikolasdehor]

Both scenarios here have been addressed by recent merged PRs:

Scenario A (native Gemini 404): PR #213 refactored providers by protocol family and PR #492 introduced model_list for flexible provider routing. The native Gemini provider should now use the correct endpoint.

Scenario B (OpenAI bridge, no API key): PR #48 added explicit provider field support in agent defaults, so you can now force provider selection:

"agents": {
  "defaults": {
    "provider": "openai",
    "model": "gemini-1.5-flash"
  }
}

If still hitting issues on the latest release, the model_list approach from #492 is the recommended path -- define your model explicitly with its provider and endpoint in the model_list config section.