This Chrome extension adds the power of large-language models (LLMs) to Overleaf through a Chrome extension.
The extension originated from GPT4Overleaf.
Unless you are a developer, this is probably your preferred option.
Just go to the extension's page in Chrome Web Store and press "Add to Chrome"
- Clone the repository
- Open Chrome and go to
chrome://extensions/ - Enable developer mode
- Click "Load unpacked" and select the repository folder
The plugin can be configured by clicking the plugin button in the Chrome toolbar.
It requires inserting an API key from OpenAI.
You also need to choose which tools you wish to enable. See the Issues section if you have problems activating the tools.
These are the tools that are currently available:
Select a text and press Alt+C to trigger the auto-complete tool.
Select a text and press Alt+I to trigger the improvement tool. The original text will be commented out and the improved text will be inserted below it.
Select a text and press Alt+A to trigger the ask tool. The original text will be deleted and the answer will be inserted in its place.
For example: "Create a table 4x3 that the first row is boldface" will be replaced with, e.g.:
\begin{tabular}{|c|c|c|}
\hline
\textbf{Column 1} & \textbf{Column 2} & \textbf{Column 3}\\
\hline
Entry 1 & Entry 2 & Entry 3\\
\hline
Entry 4 & Entry 5 & Entry 6\\
\hline
Entry 7 & Entry 8 & Entry 9\\
\hline
\end{tabular}You can then, for example:
- Write before the table: Place the following tabular inside a table environment, center it, and give the following title: "The comparison of the three approaches"
- Select the sentence and the table
- Press
Alt+Ato trigger the ask tool.
The result will be:
\begin{table}[h]
\centering
\caption{The comparison of the three approaches}
\begin{tabular}{|c|c|c|}
\hline
\textbf{Column 1} & \textbf{Column 2} & \textbf{Column 3}\\
\hline
Entry 1 & Entry 2 & Entry 3\\
\hline
Entry 4 & Entry 5 & Entry 6\\
\hline
Entry 7 & Entry 8 & Entry 9\\
\hline
\end{tabular}
\end{table}Other LLM deployments/models are supported as long as they can be accessed via OpenAI Chat Completion API. Some examples: vLLM models, LLAMA models, and easyLLM.
Use the plugin's JSON editor to change the URL and the model. See Issue #8 for further details.
If nothing happens when you use the plugin, verify that the plugin's shortcuts are not in conflict with other plugins' shortcuts. To do so, go to chrome://extensions/shortcuts.
If you encounter any problem/question, please open an issue in the project's repository.
The plugin saves its configuration locally on the user's computer. The plugin sends the API key and the selected text to OpenAI only, and only for the purpose it was made for (i.e., completing and improving text and asking GPT questions). The plugin's authors are not responsible for what OpenAI does with this data. The plugin's authors do not collect any data from the plugin's users.
If you feel advanced, you can also change the request JSON sent to OpenAI and the base URL. To do that, go to the 'Advance Configuration' component on the configuration page. By default, the value is
{
"openai": {
"url": "https://api.openai.com/v1/chat/completions",
"base": {
"n": 1,
"temperature": 0.5,
"model": "gpt-3.5-turbo"
},
"Complete": {
"max_tokens": 512,
"messages": [{
"role": "system",
"content": "You are an assistant in a Latex editor that continues the given text. No need to rewrite the given text"
}]
},
"Improve": {
"messages": [{
"role": "system",
"content": "You are an assistant in a Latex editor that improves the given text"
}]
},
"Ask": {
"messages": [{
"role": "system",
"content": "You are an assistant in a Latex editor. Answer questions without introduction/explanations"
}]
}
}
}Base is the default configuration which is overridden by the specific command configuration.