-
|
Hi, Now, I want to setup online models in pico, as long as it's free or I could use google ai studio free API key. I tried adding providers in front of "model_list":[], like so: I got: But I've seen this "providers" in some youtube tuts which works fine, although he's not using docker, and I did not see model_list[] as much as that video showed. |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments 4 replies
-
|
The error is because your config mixes the old format with the current one. Current PicoClaw config does not accept a top-level For Google AI Studio / Gemini API, use {
"model_list": [
{
"model_name": "gemini-2.5-flash",
"provider": "gemini",
"model": "gemini-2.5-flash",
"api_keys": ["YOUR_GOOGLE_AI_STUDIO_API_KEY"]
},
{
"model_name": "gemma-4-26b-a4b-it",
"provider": "gemini",
"model": "gemma-4-26b-a4b-it",
"api_keys": ["YOUR_GOOGLE_AI_STUDIO_API_KEY"]
}
],
"agents": {
"defaults": {
"model_name": "gemini-2.5-flash"
}
}
}If the Gemma model ID still fails after that, the config format is fixed but the model ID is not accepted by the Gemini API endpoint used by PicoClaw. In that case, first test with a known working Gemini model such as So the immediate fix is:
The YouTube tutorial you saw is probably for an older config version. |
Beta Was this translation helpful? Give feedback.
-
Thanks a lot for your detailed instruction. I tried this setup, with adding only one model - gemini-2.5-flash. I followed official docker installation steps. I got this: I tried both with and without : "api_base": "https://generativelanguage.googleapis.com/v1beta" in model_list. |
Beta Was this translation helpful? Give feedback.
-
|
Today I'm trying binary approach. At first I run picoclaw onboard, then picoclaw-launcher -public, then http://ip:18800, then I try to setup a model adding api key, but I get this: Maybe it's a bug, at least there's no _comment any where in config.json. |
Beta Was this translation helpful? Give feedback.
-
|
I missed one important detail earlier: adding The first Docker error is this part:
That means {
"model_list": [
{
"model_name": "gemini-2.5-flash",
"provider": "gemini",
"model": "gemini-2.5-flash",
"api_keys": ["YOUR_GOOGLE_AI_STUDIO_API_KEY"]
}
],
"agents": {
"defaults": {
"model_name": "gemini-2.5-flash"
}
}
}
The later For your container, first make sure you are editing the exact file from the log: docker compose exec picoclaw-gateway sh -lc 'grep -nE "_comment|api_key|command_allowed_remotes|model_name" /root/.picoclaw/config.json'Then:
One more version clue: current |
Beta Was this translation helpful? Give feedback.
-
|
Thank you. compose.yaml: (where /root/.picoclaw seems fixed my issue, the original was /.picoclaw) config.json: However, there're new issues: Now the status: Is webui, or "pico", one of channels? Any idea? |
Beta Was this translation helpful? Give feedback.
-
|
Yes. A couple of concrete things stand out here. First, the volume change is the right direction. The process reads the config under For curl -i http://ip:18800/health
curl -i http://ip:18800/readyIf For web search, the env var name looks wrong. Current config uses environment:
- PICOCLAW_TOOLS_WEB_PROVIDER=duckduckgo
- PICOCLAW_TOOLS_WEB_DUCKDUCKGO_ENABLED=true
- PICOCLAW_TOOLS_WEB_PREFER_NATIVE=falseOr in "tools": {
"web": {
"provider": "duckduckgo",
"prefer_native": false,
"duckduckgo": {
"enabled": true,
"max_results": 5
}
}
}
And yes, After changing env/config, restart the container or use the configured |
Beta Was this translation helpful? Give feedback.
-
|
With your help, and lots of test along with AI, I finally make it work with cli. Docker still fails, it seems docker can run gateway but not launcher. Much appreciate it. |
Beta Was this translation helpful? Give feedback.
Yes. A couple of concrete things stand out here.
First, the volume change is the right direction. The process reads the config under
/root/.picoclaw, so mounting your persistent config there is the path I would use. Mounting it at/.picoclawcan leave the process reading a different home/config directory.For
/ready: the log line only means the HTTP endpoints are listening./readyreturnsnot readyuntil startup has finished and the readiness flag is set, and it also returns 503 if a registered health check is failing. I would compare the two endpoints including status code and body:If
/healthis OK but/readyis 503, the JSON …