A plugin for GPT-3 AI assisted note taking in Logseq. Uses OpenAI's ChatGPT API `gpt-3.5-turbo` model by default.
Follow me on Twitter for updates and examples of how I use this plugin: @bsunter
To bring up the gpt popup, use the keyboard shortcut cmd+g, or select gpt from the block or slash menu.
If you are currently in a block, the plugin will use the text in the block as input to the prompt.
You can click and drag or shift+click to select multiple blocks to use as input to the prompt.
If you are not in a block, the plugin won't add any additional input text to your prompt, and will append the results of the prompt to the bottom of the page.
After selecting the prompt and generating a response, a preview of the response will be shown in the popup. You can click the Insert button or press the enter key to insert the response into the page.
You can also click the Replace button to replace the selected block with the response.
If you don't like the response, you can click the Regenerate button to generate a new response.
After you open the popup, you can write any text you want to use as the prompt.
For example you can write create flashcards based on the following text: and the plugin will generate flashcards for you to study:
There are a number of built in prompt templates that you can use to generate text. These are useful for generating outlines, summaries, and other tasks.
You can also create your own custom prompt templates.
To do this, you create a block with the prompt-template:: property. The template will be added to the list of templates in the gpt popup.
The prompt-template:: property is the name of the prompt template.
In a block nested underneath the template block, create a code block in triple backticks with the language set to prompt. The text in the code block will be used as the prompt. Make sure the code block is in its own block indented underneath the template block.
For example, you can create a template like this:
- # Student Teacher Dialog prompt-template:: Student Teacher Dialog - ```prompt Rewrite text as a dialog between a teacher and a student: ```
To replace the selected block with the generated text, click the Replace button.
If you don't like the output of the prompt, you can click the Regenerate button to generate a new response. Sometimes the first response is not the best, but the second or third response can be better.
Type /gpt-block in a block or select gpt-block from the block menu.
gpt-block will send the block to OpenAI's GPT-3 API and append the response underneath the block.
Type /gpt-page in a block or select gpt-page from the block menu.
gpt-page will send the entire page to OpenAI's GPT-3 API and append the response to the bottom of the page.
Transcribe audio files to text using the Whisper API.
Type /whisper in a block or select whisper from the block menu.
Supported formats are:
m4a, mp3, mp4, mpeg, mpga, wav, webm
Note, that the Logseq mobile app audio recorder uses .aac which is not supported by Whisper. I recommend using a separate audio recorder app to record audio files and then uploading them to Logseq. For example, I use the Voice Memos app on iOS and share that file to the Logseq mobile app.
Generate images with DALL-E
Type /dalle in a block or select dalle from the block menu.
This will generate an image using the DALL-E model, save the image to the assets folder, and insert the image into the block.
You can click and drag or shift+click to select multiple blocks to use as input to the prompt.
You can adjust the chatPrompt setting to adjust how ChatGPT should respond to your input. By default, the setting is set to Do not refer to yourself in your answers. Do not say as an AI language model... to prevent the model from including unnecessary text in the response.
You can add guidance such as "respond in chinese" or "respond in spanish" to the prompt to get the model to respond in a different language.
This setting is only used when the model is set to gpt-3.5-turbo.
For people who uses reverse-proxy server to use OpenAI service ,you can set the chatCompletionEndpoint to your reverse-proxy endpoint , the default configuration for this is http://api.openai.com/v1.
WARNING: To use those reverse-proxy endpoints , you should always keep your data and privacy safe !!!
Allows you to inject a prefix into the GPT-3 output before it is inserted into the block, such as a [[gpt3]] tag or markdown formatting like > for a blockquote. This is useful for identifying blocks that were generated by GPT-3.
Use the Inject Prefix options in the setting to set the prefix. You can add a space or \n newline to separate the prefix from the generated text.
logseq-plugin-gpt3-openai allows users to generate human-like text using GPT-3 within the LogSeq editor.
Write a GPT-3 command in a block, then run the open /gpt command via the slash or block menu. The plugin will generate a GPT-3 response using the OpenAI API and insert it below. It removes leading and trailing whitespace from the prompt.
-
You need to get an OpenAI API Key here and add the key in the plugin settings.
-
After the free welcome credit expires you will need to provide a credit card to OpenAI.
-
Note that having a paid ChatGPT account is not the same as having a paid account for OpenAI's API.
-
The default model is
gpt-3.5-turbobut you can change it totext-davinci-003for GPT-3 orgpt-4(if you have access)in the plugin settings underopenAICompletionEngine. See the OpenAI docs for available models. -
Make sure you read OpenAI's usage guidelines and avoid generating certain types of content.
-
Download the plugin in the Logseq marketplace by searching for
gpt3oropenai.
GPT-3 has limitations. It sometimes produces output that is subtly wrong or misleading. Don't rely on its output without verifying it yourself. Use it with caution.
-
Generate code to do a given task
-
Correct grammar
-
Translate into other languages
-
Classification and keyword tagging of text
-
Generate lists of given topics
List 10 top selling science fiction books
-
Write about a given topic
Write a tagline for an ice cream shop.
-
Answer Questions
Q: How does a telescope work?
See this article for a good overview.
Your free trial is over, or you've run out of tokens. You can refill your tokens here.
OpenAI has limits on how often you can call them. If you get this error, you'll need to wait a bit before trying again. See this article for more info on the rate limits. You can call it faster if you have a paid account.
You may have mistyped the model name, or need to generate a new API key, if you're upgrading to a paid account from a free account, or using a beta model that is not available to the public.
This error doesn't cause any issues besides the error message in the console. It's a known issue with the OpenAI API. See this issue for more details. I'm working on a PR to their library to support browser usage. Ignore this error for now.
]
-
Open the developer tools (Menu -> View -> Toggle Developer tools)
-
Check the console logs for error messages.
-
See if you can figure out the error on your own. Maybe you had a network issue if it says something like "timed out." Sometimes the OpenAI API has issues. You also have a limited number of tokens, so you may run out and need to refill.
-
If you can't figure it out based on the error message and it doesn't go away. Make an issue on GitHub.
An API key from OpenAI Click here to get one
pnpm i
- Add Tests
pnpm run build
This enables the local dev server with hot reloading, via the logseq vite plugin.
pnpm i
pnpm run dev
First run pnpm i and pnpm run build
Open LogSeq
Go to Settings > Turn on Developer Mode
This will bring up the "Plugins" entry in three dots more menu list on the top right of the header bar. Go to Plugins page, and you will get a button with the Load unpacked plugin label. Select the root folder of this plugin repo.
Make sure you add your OpenAI Key
Creates a build using semantic release when a commit is pushed with a smart commit message.
- LogSeq - Privacy-first, open-source knowledge base that works on top of local plain-text Markdown and Org-mode files.
- Vite - Next Generation Frontend Tooling
Do you have a bug or idea? I would love to hear from you! Open a GitHub issue here.
PRs welcome. Open an issue to discuss first if possible.
- @briansunter - Author
