# Deployment Requirements
In this section, we will cover the requirements for deploying the project.


## LLM(Large Language Model) and Embedding Model
* A saas LLM model like OpenAI API or self-hosted LLM model with requirements:
  * Smarter than GPT-3.5
  * Provide openai-like API
* Embedding model: AutoFlow needs an embedding model to translate the text into vectors. You can use the following:
  * OpenAI-like embedding model
  * Cohere embedding model
  * ZhipuAI embedding model
  * You can also use the [Jina AI API](https://jina.ai/) for this purpose. It is free for 1M tokens.
* (Optional) Reranker. You can use the [Jina AI API](https://jina.ai/) for this purpose. It is free for 1M tokens.


## TiDB
* With [TiDB Serverless](https://pingcap.com/ai) account, you can setup a TiDB cluster with Vector Search enabled. Free quota is available for 1M RU per month.
* You can also use a self-hosted TiDB cluster(>v8.4) with Vector Search enabled, but please note it will require TiFlash enabled for Vector Search.


## Hardware

### If you are using a Cloud TiDB and SaaS LLM
You can use any of the following web hosting services to deploy the project:
* Cloud server providers like [AWS](https://aws.amazon.com/), [Google Cloud](https://cloud.google.com/), [Azure](https://azure.microsoft.com/), etc.
* Or your own server.

We suggest the following configuration for the server:

| Name                 | Value            |
|----------------------|------------------|
| CPU                  | 4 vCPUs          |
| Memory               | 8 GB RAM         |
| Disk                 | 200 GB SSD       |
| Number of servers    | 1                |


### If you are using a self-hosted TiDB and self-hosted LLM
If you use a self-hosted TiDB and self-hosted LLM, you need a powerful server to handle the load. We suggest the following configuration for the server:

| Name                 | Value            |
|----------------------|------------------|
| CPU                  | 32 vCPUs         |
| Memory               | 64 GB RAM        |
| Disk                 | 500 GB SSD       |
| GPU                  | 1 x NVIDIA A100  |
| Number of servers    | 1                |

GPU here is used for the LLM model, you can use any other GPU model that can be used for the LLM model which has capability more than gpt-3.5.