如何配置本地私有模型？本地运行是否有像dashboard一样的可视化界面

[sunhao-Y]

1、如何配置本地私有模型？即自己指定base url为localhost，而不是各种llm供应商
2、本地运行是否有像dashboard一样的可视化界面

[SuyashR25]

You can add these lines at the start of your code (after importing litellm, os, dotenv)

litellm.api_base = "YOUR_LOCAL_ENDPOINT_URL"
litellm.api_key = "YOUR_LOCAL_KEY"
litellm.drop_params = True
os.environ["SSL_VERIFY"] = "0"

[wangjiawen2013]

如果想指定其它大模型厂商的API呢，比如qwen, deepseek

[KylinMountain]

1. Local / private model (localhost base URL) — point PageIndex at any OpenAI-compatible endpoint via LiteLLM. In pageindex/config.yaml:

model: "openai/<your-model-name>"

and in .env:

OPENAI_API_BASE=http://localhost:8000/v1
OPENAI_API_KEY=any-placeholder   # required even if your server has no auth

For a local Ollama server, use model: "ollama/llama3" instead (LiteLLM defaults to http://localhost:11434, override with OLLAMA_API_BASE).

@wangjiawen2013 for other vendors just use the LiteLLM model name — e.g. deepseek/deepseek-chat, or Qwen via dashscope/qwen-max (or an openai/-compatible endpoint). Full list: https://docs.litellm.ai/docs/providers

2. Dashboard / visual UI — the open-source repo doesn't ship a UI; indexing produces a structured JSON tree (pretty-print it with the print_tree() helper in pageindex/utils.py, or open the JSON in any viewer). For an interactive experience, the hosted Chat Platform is at https://chat.pageindex.ai.

Heads-up: several steps expect valid JSON from the model, so very small / non-strict local models (e.g. some DeepSeek setups — see #326) can trip the parser; we're hardening this in #333.

Closing as answered — reopen anytime if something's unclear!