Inspiration
The inspiration from this mainly occurs from existing popular chrome extensions like Sider, Monica... I tried searching for similar one for perplexity alone as it's quite good for it's searching capabilities. So I stumbled upon the official one from perplexity that is using popup, instead of siderbar experience for an extension which I think is better and it's UI and UX is not that intuitive.
What it does
You can access this as an extension within your browser and enter your Perplexity API key and get access to all the sonar models that perplexity offers and converse with it according to the model that you like and the kind of answer that you expect from each models.
How I built it
I built it with these follow Tech Stack:
- Frontend: React, TypeScript, Tailwind CSS
- UI Components: Radix UI
- Markdown Rendering: react-markdown with remark-gfm and rehype-raw
- Build Tools: Vite, esbuild
- Extension Framework: Chrome Extension Manifest V3
- AI API: Perplexity Sonar API
Challenges I ran into
This was my second try at building a chrome extension before this, I tried building a chrome extension before AI buzz, I gave it up last time cause I couldn't build it and didn't have much motivation to continue. But for this project I used tools like github copilot and cursor with claude 3.7 models and it was amazing but also regularly introduce a lot of bugs and errors, It took me hours sometimes of back and forth and fighting with it and threatning claude that I might loose the hackathon and sometimes it worked sometimes it didn't so I had to do it the old way of googling and digging through stackoverflow and/or docs and fixing some of those bugs.
Also all the Sonar Models seems to be trained to provide detailed answers, even for simple conversational message like Hi, it was telling what it was used for, how it's use and all that yapping. Maybe I could tweak it's behaviour and how it responds through system message and adjusting temperature, top_p or maybe some other attributes but I wish perplexity themselves set the most suitable and better default values for these.
Accomplishments that I'm proud of
This is my first ever successful try at chrome extension and I'm happy about how it turned out.
What I learned
Perplexity has been developing these sonar models on top of base Llama LLM and the moat of building product on top of these LLM's is more than building the underlying LLM's. It's all about the delightful experience that the customers want no matter which stack it's built on at the end of the day customers want what they want and Perplexity's Sonar models are quite good at searching, reasoning and deep research related tasks so instead of visiting perplexity.ai and searching their query, users can directly click on extension that opens on their sidebar and can directly ask it and select the texts on any webpage for context.
What's next for Perplexity Browser Companion
- Possible give access to webpage that user is currently on so sonar models can get more context and also can be helpful for users to ask the questions related to the webpage.
- Support artifacts that can be displayed right within the response from AI.
- Publish to chrome and edge webstore and improve it further based on user feedback.
Built With
- radix
- react
- react-markdown
- tailwind
- typescript
Log in or sign up for Devpost to join the conversation.