khoj/documentation/docs/advanced/litellm.md

LiteLLM

:::info This is only helpful for self-hosted users. If you're using Khoj Cloud, you're limited to our first-party models. :::

:::info Khoj natively supports local LLMs available on HuggingFace in GGUF format. Using an OpenAI API proxy with Khoj maybe useful for ease of setup, trying new models or using commercial LLMs via API. :::

LiteLLM exposes an OpenAI compatible API that proxies requests to other LLM API services. This provides a standardized API to interact with both open-source and commercial LLMs.

Using LiteLLM with Khoj makes it possible to turn any LLM behind an API into your personal AI agent.

Setup

1. Install LiteLLM

   pip install litellm[proxy]
   

2. Start LiteLLM and use Mistral tiny via Mistral API

   export MISTRAL_API_KEY=<MISTRAL_API_KEY>
   litellm --model mistral/mistral-tiny --drop_params
   

3. Create a new OpenAI Processor Conversation Config on your Khoj admin panel
   • Name: proxy-name
   • Api Key: any string
   • Api Base Url: URL of your Openai Proxy API
4. Create a new Chat Model Option on your Khoj admin panel.
   • Name: llama3.1 (replace with the name of your local model)
   • Model Type: Openai
   • Openai Config: <the proxy config you created in step 3>
   • Max prompt size: 20000 (replace with the max prompt size of your model)
   • Tokenizer: Do not set for OpenAI, Mistral, Llama3 based models
5. Create a new Server Chat Setting on your Khoj admin panel
   • Default model: <name of chat model option you created in step 4>
   • Summarizer model: <name of chat model option you created in step 4>
6. Go to your config and select the model you just created in the chat model dropdown.