AI Chat

SAL (Sigasi AI Layer, in case you’re wondering) is the name of the integrated AI chatbot in Sigasi Visual HDL. There are three ways to get a conversation with SAL started:

Open Chat with SAL in the Sigasi view
Toggle the Command Palette (Ctrl-Shift-P) and choose Sigasi Visual HDL: Focus on Chat with SAL View
Right-click anywhere in the editor and select SAL > Explain This Code

Note: Through SAL, you can connect to a remote model using the OpenAI API, such as OpenAI’s GPT 4 model, or a local AI model of your choice via LM Studio.

Configuring your AI model

SAL is configured using up to four environment variables. Be sure to set them before starting Sigasi Visual HDL, so they get picked up correctly. The easiest way to get started it by connecting to the OpenAI servers, as detailed below. If you prefer to use a model made by another company, or you’re working on an airgapped machine, you’ll have to set up a local model.

Connecting to remote OpenAI services

If you have a working internet connection and an OpenAI API key, configuring the backend for SAL is as easy as setting the environment variable SIGASI_AI_API_KEY to your API key. By default, this will use the GPT 3.5 Turbo model.

export SIGASI_AI_API_KEY="your-openai-api-key"

You can use a different model by setting the SIGASI_AI_MODEL to e.g. gpt-4-turbo.

export SIGASI_AI_MODEL="gpt-4-turbo"           # with the GPT 4 model
export SIGASI_AI_API_KEY="your-openai-api-key" # with this API key

For more details on setting environment variables, refer to this guide.

Connecting to a local model

This guide will help you use LM Studio to host a local Large Language Model (LLM) to work with SAL. Currently, SAL supports the OpenAI integration API, and any deployed server utilizing this API can interface with SAL. You need to set the correct URL endpoint and model name, and optionally provide the API key if required by the endpoint.

Download the latest version of LM Studio .
Open the LM models search engine by clicking this search icon from the top left pane.
Search for an LLM of your choice, e.g., DeepSeek Coder V2 Lite, and click download.
Once the download is over, a pop-up window will show up offering to load the model directly.
LM Studio automatically switches to Chat mode once the model is loaded. Switch to developer mode.
Double-check that the DeepSeek model is loaded and displayed on the “Loaded models” tab. Click Start Server (Ctrl + R).
After a successful run, the following messages should be displayed in the “Developer Logs” tab.
```
2024-11-26 15:32:24 [INFO] Server started.
2024-11-26 15:32:24 [INFO] Just-in-time model loading active.
```
(Optional) Configure local server parameters, such as the Context overflow policy, the server port, and Cross-Origin-Resource-Sharing (CORS).
Set the following environment variables:
- SIGASI_AI_API_URL
- SIGASI_AI_MODEL
- SIGASI_AI_API_KEY
  to the following values respectively:
- The specified URL as shown in this image
- The specified model as shown in this image
- The string value: “lm-studio”
  In this example:
```
export SIGASI_AI_MODEL="deepseek-coder-v2-lite-instruct"
export SIGASI_AI_API_URL="http://127.0.0.1:1234/v1/"
export SIGASI_AI_API_KEY="lm-studio"
```
Launch Sigasi Visual HDL and start a conversation using your configured local LLM.