Note
Pixlie is currently in beta. We value your feedback as we progress towards a stable release.
AI powered knowledge graphs for semantically accurate insights. From online search to private document search to semantic search in your apps.
Point Pixlie at any website and describe what you need - from pricing data to project details. Our open-source intelligent crawler does the rest, no coding required.
- Objective driven projects to create knowledge graphs ("Track companies on Indian stock exchanges")
- Use Anthropic Claude (other models' support coming soon) with your own API keys
- Use Brave Search API for web search with your own API keys
- Built-in simple web crawler
- Runs on your laptop or on the cloud
- Collaborate with your team (paid feature)
- Extract entities like people, places, dates, events, etc. (some paid features)
- Search your knowledge graph
- Use as your default search engine
- Simple to download and install
- Completely UI driven
Watch this video to see Pixlie in action:
Choose your objective - what do you want to discover with Pixlie?:
Information will instantly be populated. Let Pixlie search a while and see...
A list of links discovered based on your objective:
A list of domains (as many links may come from one domain):
The content extracted from these links:
Search through the results for specifics:
At the moment Pixlie can be used on your laptop for personal web research (desktop app coming soon!). Pixlie is open source, with the option of enterprise licensing for those who want to utilize Pixlie's search and knowledge graph within their product.
Pixlie is open source and is under active development. Our work is done in public, please star this repository, it means a lot to us.
If you want to develop on Pixlie, please see the DEVELOP.md file.
If you want to use Pixlie, please see the USE.md file.
Pixlie has a REST API that you can use to interact with the graph. We use Bruno to document the API.
You can find the API spec in the rest_api directory.
If you want to be notified when Pixlie is ready for use, please subscribe to our insights newsletter.
Join the discussion or get support on Discord.
- Pixlie is licensed under the GNU General Public License version 3.0
- See the LICENSE file for details
Yes, Pixlie is an alternative to using vector databases. Vector databases are good for storing and querying semantic data, but they do not model the underlying data accurately. In Pixlie, we use LLMs to classify individual pieces of semantically meaningful data. Each individual entity, such as a person, place, date, event, etc., is stored separately in the graph, along with its relationships to other entities. This makes our graph based approach better where accuracy is important.
Link Nodes are the URLs we discover using Brave & AI based on the Objective. When each Link Node is processed, we store domains for these Links in our database as separate nodes - one node for every domain.
We connect Domain Nodes to their Link Nodes in the following way:
Domain —OwnerOf—> Link(s)
Link —BelongsTo—> a Domain
As a part of the processing of Link Nodes, we then fetch their raw HTML content. Once we successfully fetch the content, we create a WebPage Node to store that content and then connect it to the Link Node:
Link —PathOf—> WebPage
and vice-versa
WebPage —ContentOf—> Link