magic-chart-js
v0.0.4
Published
Wrapper around Chart.js that uses AI to generate configuration.
Downloads
2
Readme
✨ Magic Chart.js
A proof of concept wrapper around Chart.js that uses an LLM to generate the chart, based on the data you want to display.
Simply pass in results
to the chart and the LLM will decide how to display them! You can also pass in instructions via the prompt
, if you have something in mind.
This is currently very much a proof of concept!
Setting up
Start by installing this library:
npm i magic-chart-js
To power the LLM features, you'll have to set up Relevance AI.
To do this, you'll need to install the Relevance AI SDK:
npm i -g @relevanceai/chain
relevance login
Create a chains folder at the root of your project, and create a file called magic-chart-js.ts
(can be .js
if that's your jam!). Inside that folder, add the LLM chain which powers the magic.
Copy this template and save!
Now deploy it:
relevance deploy
You're off to the races!
TIP: You can customise this chain as much as you like! For example, you might want to tweak the prompts or set it up to use different LLMs. However, our we expect it to always receive results
and prompt
as params, as well as return a chartConfiguration
output variable. For more information, check out Relevance AI docs.
Using it
Start by creating a client, using your Relevance AI details:
import MagicChart from 'magic-chart-js';
const chart = new MagicChart({
region: RELEVANCE_AI_REGION,
project: RELEVANCE_AI_PROJECT,
});
Then call our summonChart
method to render the chart.
chart.summonChart({
item: document.getElementById('my-canvas'), // must be a canvas element
results: [...], // an array of any data
prompt: 'Make it a pie chart please!', // optional
});
item
This must be a canvas
HTML element, where the chart will be rendered by Chart.js
.
results
This can be an array of any data. The LLM will decide how to display it!
NOTE: The more results you feed in, the larger your prompt sent to the LLM. This means more $$$ and also, the potential to break context limits. I recommend just sending a controlled amount, like 5-20 items at a time!
prompt
Optionally, you can send in a request! You could even let your users do this!
License
MIT © danieljpalmer