browser knowledge graph tool runs fully in browser
The open-source project edgextract enables the creation of a browser knowledge graph directly in the client, eliminating any need for a backend server or API keys. Built with Rust, compiled to WebAssembly and accelerated by WebGPU, the entire pipeline executes inside the user's browser.
browser knowledge graph
Key capabilities include automatic extraction of entities and their relationships from raw text. Instead of asking a large language model to generate a bulk JSON payload that may contain errors, edgextract leverages the 0.8 billion-parameter Tev1 model from Together AI to evaluate each potential relationship with a simple yes/no query. Each edge receives a confidence score ranging from 0 to 1, allowing developers to set a threshold that automatically accepts, discards, or flags a relation for human review.
The tool visualises the resulting graph with D3, and each edge can be traced back to the original sentence that produced it, providing full transparency.
Performance and accuracy
Benchmarks show the browser-based engine runs up to 18 times faster than a comparable locally-run chat LLM. On the standard CoNLL04 dataset, the number of incorrect relationships dropped from 504 to 218, demonstrating a substantial improvement in precision.
A notable privacy advantage is that all data stays on the client device; no text or extracted information is sent to the cloud.
References
These external sources were used to verify the article and provide deeper context.
Conclusion
edgextract delivers a server-less, privacy-preserving solution for building knowledge graphs in the browser, combining high speed, configurable confidence thresholds, and open-source accessibility.
References
