[bug]Error: LLM request failed. Check your API key and model configuration in Settings.

[gxlqssjf]

win11配置LM studio本地模型时,使用时报错如下:
Error: LLM request failed. Check your API key and model configuration in Settings.
配置文件如下:

OpenFang Agent OS configuration

See https://github.com/RightNow-AI/openfang for documentation

api_listen = "127.0.0.1:4200"

[default_model]
provider = "openai"
model = "qwen2.5-1.5b-instruct"
base_url = "http://localhost:1234/v1"
api_key_env = ""

[memory]
decay_rate = 0.05

[yrwy]

这种傻逼玩意直接删了好了 什么玩意 居然要配置Embedding 还不能关闭
config.toml

[default_model]
api_key_env = "CPA_API_KEY"
model = "qwen3.5-plus"
provider = "cpa"

[provider_urls]
cpa = "http://127.0.0.1:8317/v1"

custom_models.json
[
{
"id": "qwen3.5-plus",
"display_name": "qwen3.5-plus",
"provider": "cpa",
"tier": "custom",
"context_window": 1000000,
"max_output_tokens": 65536,
"input_cost_per_m": 0.0,
"output_cost_per_m": 0.0,
"supports_tools": true,
"supports_vision": true,
"supports_streaming": true,
"aliases": []
}
]

secrets.env

CPA_API_KEY=fuckuall@@@

配置自定义模型需要这样 你可以参考下 发完这个就去删了 一个用vibecoding出来的垃圾 顶着一个rust的帽子
支持的模型都是上古的

[jaberjaber23]

Fixed the empty api_key_env handling. When api_key_env is set to an empty string (meaning no API key needed for local inference), the kernel was calling std::env::var("") which always fails. Now it checks if api_key_env is empty first and skips the env lookup. Also fixed the localhost auth issue so the dashboard works without api_key when accessed locally.