Underlay, basis for SAFE knowledge graph?


One of the co-founders of Metaweb/Freebase, which was later sold to Google and turned into Google Knowledge Graph, is now starting a new project to create an open knowledge graph, essentially what freebase was supposed to be originally. The project is called the Underlay.

The website doesn’t have that much info, but there’s a video/article here

They’re seeking partners for the project. SAFE could be a good location to host such a graph and perhaps it could be a good idea for MaidSafe to talk to them to discuss data format, integration with Solid etc to ensure that it be hosted and run smoothly on the SAFE Network once it launches.

They’re currently planning to use IPFS as the main host it looks like.

Fuzzy logic usage
My vision for a universal community and database website

Good post. Interestingly their ‘partners’ link is an MIT email address (where Tim is, though he’s now on sabbatical with Inrupt). If I were you I’d go over to Solid chat or forum and ask if he/they know about it and whether it is SemWeb related. Seems odd they don’t mention Linked Data, so worth poking.

I agree it would be worth @Maidsafe making contact with Underlay too. cc @dugcampbell


Thanks for the links @intrz very interesting video interview.

The Underlay appears to lend itself towards the way Safe will work (Append Only, history, immutable and consistent data representation) so may be worth leaving the door open to. Timely to consider given that the team is considering ways to bake RDF, linked data in.

From: “The Future Of Knowledge” at https://underlay.pubpub.org/

Comparisons to Other Public Knowledge

The Semantic Web is a public machine-readable collection of data with a certain format posted to any public web page. Like the Underlay, the Semantic Web contains has public assertions of relationships between entities. However, there is little provenance for these assertions beyond what is implied by the page that published the data. Assertions are made and discovered ad hoc. All assertions are editable by the page owners, with no requirement to maintain history, immutability or consistency. Different semantics of relationships and entities are used by different publishers, who bear the burden of publishing their own data.
Linked data (or linked open data) is a promising way of representing information by breaking it into a set of assertions, each of which is a triplet of subject and predicate joined by a relationship. The terms ideally link to a Web site or service that defines it. Linked data can represent knowledge without over-simplifying it to fit a set of expected fields, or committing to a schema beforehand. It also enables stitching together information from different collections. Unlike the Underlay, linked data is just a way of representing information. It is not append-only; data points are often not timestamped and can be edited in place. Including provenance for each triple is not standardized and can be complex. The Underlay could make use of the format, and work with existing linked data collections.

Also worth reading “The Underlay Project: patterns for crystallizing knowledge” also at above link.


I’m not sure that just because they don’t say the word it doesn’t qualify as such, especially if we consider Freebase was a triplestore.

They seem to want to store qualified statements, but that can be expressed with triples as well, with the statement itself is a separate entity and “according to”, “in year”, etc added as qualifiers.


Hi @intrz,

Thank you for linking to this project! Although a bit overdue, I contacted the authors and asked several questions about the technologies they aim to base it on. First of all, I asked if they plan to use RDF, Linked Data, and SPARQL.

Second, there’s an interesting bit in the project description – seems that they consider distributing the knowledge graph using IPFS. So I asked if they want to make the protocol closely integrated with the IPFS tech stack, or if it is interchangable. It would be very interesting to store it on the SAFE Network in form of Linked Data: with the RDF/querying capabilities available in the core API any app can fetch and utilise data from the knowledge graph. Same applies to Solid, actually.

I’ll update this topic once I get the answers. I also reposted to the Solid community forum:


Something like this?