Hypertext is a medium for thought and communication. The advent of the web simplified the more complex hypermedia systems developed from the 1960s to the 1980s. The web's emphasis on networking and decentralization democratized access to information, propelling it to the forefront of the hypertext revolution and securing its position as the dominant force.

However, despite its foundational promise, the web has fallen short of its potential as the ultimate medium for thinking and communication. Centralized platforms like Twitter and Instagram dominate, driven by central planning that stifles digital societies’ dynamic and collaborative nature. This centralization impedes innovation and collaboration, undermining the potential of digital media.

Humans need better collaboration and thought tools to solve ever more complex problems. However, the goal of most web content today, basically social media or click-bait media, is to generate advertising by competing for our attention through emotional manipulation.

More than ever, the public sphere has become a platform for advertising instead of a platform for constructive dialogue and critical debate. Conversations are fragmented, divisive, and plagued with disinformation, leading us to polarization.

Seed Hypermedia emerges as a solution to fix collaboration on the web within a framework of decentralization and networking. Drawing inspiration from pioneering hypermedia systems like Augment and Xanadu and projects like the Open Hypermedia System from the University of Southampton [7], Seed Hypermedia bridges the gap between traditional hypermedia and the modern web.

Seed Hypermedia marries cryptographic signatures with a promise from the past: a publishing system with fine-grained linking and transclusion capabilities. By integrating decentralized technology with content distribution, Seed Hypermedia empowers creators to collaborate while retaining authorship rights, offering their audiences and communities social media interactivity within a self-curated environment. To achieve these goals, we will follow the principles below:

1.3.1 Open Source Hypermedia System for the Web. Our codebase must be open source if we aim to amplify human knowledge through meaningful collaboration. Open source software empowers users to modify and improve the software, bootstrapping tools that best support their collaborative processes and information workflows. This approach ensures transparency, trust, and security, essential for systems that manage and disseminate human knowledge.

1.3.2 Open collaboration for knowledge communities. People's collaboration should focus on shared goals to build comprehensive knowledge bases needed to address collective challenges rather than an empty competition for consumers attention. By harnessing collective intelligence, distributed communities can effectively pool resources and expertise, creating an environment where innovative solutions can emerge from united efforts.

1.3.3 Decentralization. Detaching identity from third-party servers ensures user autonomy and privacy, promoting a more secure and democratic digital identity framework. Seed Hypermedia prioritizes dissociating user identity from centralized servers in line with the concept that servers should serve merely as infrastructure. By empowering users to control their own identities and authentication mechanisms, we foster a more resilient and inclusive digital ecosystem where individuals have greater autonomy over their personal data and interactions.

1.3.4 Peer-to-Peer Content Distribution. Direct and open content distribution in a peer-to-peer fashion ensures resilience and decentralization, reducing reliance on centralized servers. Together with decentralized identities, we eliminate the problem of central authority controlling access to information.

1.3.5 Respect for Copyrights. Publishing should respect copyrights, balancing authors’ rights with open access and collaboration principles. Seed Hypermedia takes a proactive approach to copyright respect by implementing mechanisms for transparent attribution and licensing of content. By facilitating clear ownership and permissions management, Seed Hypermedia ensures that creators’ rights are upheld while fostering a collaborative culture within the digital ecosystem.

1.3.6 Permanent Information. Information should be permanent, with robust versioning systems to maintain the integrity and history of digital content. Embracing the Long Now concept, which encourages thinking on timescales of centuries or even millennia, Seed Hypermedia aims to ensure that digital content remains accessible and meaningful across generations. By preserving information durably and timelessly, we contribute to a more enduring and inclusive digital heritage for future civilizations.

Seed Hypermedia documents are one kind of Open Hypertext Document composed of various content blocks1 that can include text, multimedia elements, external web embeds, Embeds and other Hypermedia documents. Each block has a unique identifier that remains constant even if the document is moved or updated. This identifier system facilitates character-level referencing within documents, enhancing collaboration and discussion.

Documents are structured as lists of blocks, with blocks capable of nesting other blocks, forming a hierarchical structure. This organization helps manage complexity, supports cognitive processes, and aligns with human capabilities, enhancing efficiency and comprehension.

Every entity in a document is addressable. URLs can point to a whole document, block, or even a piece of a block. This fine-grained linking enables fair content reuse, allowing documents to bring in pieces from other documents while keeping the attribution and original context and maintaining a bi-directional link between the two. These are commonly called ‘transclusions [12, preface, p.5, fn.2] in hypermedia systems, but Seed Hypermedia uses the term Embeds. This is different from copying. A copy isn't inherently linked to its original and can be changed in the duplicate version.

Unlike HTML and other related formats, links in a Seed Hypermedia Document have a designated space and are not "sprinkled" all over the text. Instead, there is a global map of named links for the whole document, and every time a link is used in the content, it is used as a reference from this global map. This is very similar to how scientific papers have a separate reference section.

When reading a document, a user can access backlinks — mentions from other documents of their linked-data graph pointing to this document. From a side panel, the user can view different documents at the same time.

Every publication is content-addressable and replicated over the network, making broken links less possible, as any peer on the network can become a content provider. This is a crucial difference from what we're used to with links on the Web, which are location-based and prone to disappear.

This approach is based on the web of names system where every content has a convenient and easy name, a UUID that allows us to track content that changes with time, and the hashes (SHA1) that allow total precision in managing content[1].

The document hierarchy in Seed Hypermedia enables the implementation of innovative viewspecs, which provide flexible and powerful ways to navigate and manage complex documents. Inspired by Doug Engelbart's Augment system [6], viewspecs allow users to manipulate the display of document contents based on various parameters, facilitating better comprehension and organization. Key features include bird's eye view, clipping, truncation, and outline views. These tools enable users to decide how to view a document, such as showing only the first sentence of each paragraph, collapsing or expanding sections, and applying content filters. This approach allows users to gain a high-level overview and then zoom in on specific areas of interest, enhancing their ability to study, modify, and understand the document's structure and content effectively.

Additionally, viewspecs can be embedded into links, allowing different document views to seamlessly incorporate into other documents. This flexibility not only aids in better content management but also promotes a dynamic and interactive hypermedia experience by allowing users to tailor the visibility and structure of information to their preferences.

While documents serve as the primary medium for content creation, offering a solid and structured way of sharing knowledge, we also recognize the need for spontaneous and ephemeral conversational content. This is where comments come into play. Comments allow for dynamic and quick exchanges of ideas, serving as a space for academic discourse and critical thinking. They are the lifeblood of intellectual conversation, facilitating ongoing dialogue and fostering deeper understanding.

Our hypermedia system is designed to enhance open collaborative and editorial workflows through structured, version-controlled documents, deep linking, and dynamic, fine-grained permissions. Cryptographic signatures and peer-to-peer distribution allow authors to maintain sovereignty over their contributions, promoting an environment conducive to networked improvement communities.

2.2.1 Collaborative Editing. Collaborative editing is facilitated by allowing any account to introduce and declare ownership of a new branch. The system's dynamic and fine-grained permissions allow precise control over who can directly edit documents or suggest changes. These permissions can be adjusted dynamically, providing flexibility and control over collaborative processes. The permissions are engineered cryptographic allow lists and capabilities-based security. Built-in features such as decentralized identity and version control enable users to see how the document has evolved and how every user has contributed to the different versions.

2.2.2 Suggested Changes. Suggested changes allow users to propose modifications to documents they do not have permission to edit or prefer to suggest changes even if they have the necessary permissions. They do not alter the document's official version but are presented to others as recommendations. This mechanism encourages community input and peer review, fostering an environment where improvements can be discussed and refined before being officially adopted.

2.2.3 Branching alternative perspectives. Branching or forking a document allows users to create independent versions of documents when changes are already part of another document. This capability supports innovation and experimentation by enabling users to explore new directions or improvements without impacting the original document. It not only allows for the introduction of new perspectives into collaborative projects but also fosters diverse contributions. If a fork develops content that could enhance the original document, users can merge this content back into the original.

2.2.4 Conversations and Discussions. Our system also supports argumentative discussions by allowing users to suggest changes and engage in debates about the content. These discussions can occur within the context of suggested changes, where users can propose modifications and provide arguments for or against them. This encourages a thorough review process and critical thinking, as users must justify their suggestions and consider alternative viewpoints.

2.2.5 Knowledge Communities. Seed Hypermedia's networked infrastructure fosters collaboration among diverse groups from different knowledge fields, work groups, and corporations. This interoperability between various domains enhances our ability to find innovative solutions to increasingly complex challenges.

By establishing knowledge communities that engage in collaborative dialogue within a distributed repository of documents, we facilitate the emergence of structured knowledge unique to each field. Leveraging open-source software enables these communities to continuously refine and enhance their collaboration tools, significantly accelerating the achievement of their collective objectives.

In our decentralized hypermedia system, identity management ensures secure, verifiable user interactions. Instead of traditional centralized identity mechanisms, our system relies on cryptographic keys to establish and verify user identities.

3.1.1 Public Key Infrastructure. At the core of our identity system is the use of public-key cryptography, which is fundamental to enhancing security and user autonomy within our decentralized framework. Users are empowered with a unique, twelve-word ‘seed phrase’ that establishes a trusted record, while the corresponding private key acts as a personal signature. This private key is used to sign changes and actions, offering a robust mechanism for authentication and non-repudiation.

The public key, derived from this seed phrase, is broadcasted to enable other users within the system to verify the authenticity of actions performed by the account holder. All interactions, changes, and data within the system are cryptographically signed with the user's private key, ensuring that they are tamper-proof and verifiable.

This method not only eliminates the need for traditional password-based authentication, which is vulnerable to various security threats but also significantly enhances the security environment for user interactions, adhering to the principles of user control inherent in decentralized systems.

3.1.2 Key Management Challenges. Managing cryptographic keys in a decentralized environment presents unique challenges. Key revocation, for instance, is problematic because invalidating a key also invalidates all content previously signed with that key. This would be unacceptable in a system where content is meant to be permanent and reusable. Our solution involves a careful balance between usability and security, where users are encouraged to securely store and back up their keys. Additionally, we support key rotation to allow users to replace compromised keys without losing their entire identity. This approach requires users to rebuild trust through social proofs and other mechanisms, which, while cumbersome, are necessary for maintaining security.

3.1.3 Web of Trust and Address Book. To prove their identity, users can publish cryptographically signed claims on various third-party platforms (social media, forums, DNS, HTTPS). In addition to that, users can vouch for the veracity of the identity of other users by broadcasting signed claims on the network [13]. The strength of one's identity is proportional to various proofs it has, but overall it is subjective to each individual counterparty: e.g. Alice can decide to trust Bob because she knows him personally, even if Bob does not have any signed proofs at all.

Each user maintains an Address Book that aggregates various identities under an arbitrary name, similar to the address book on a mobile phone. This address book is public, helping to spread trust across the network.

The system employs IPFS [2] for immutable data storage, leveraging content-addressable identifiers (CIDs) and CRDTs to manage version-controlled documents. Document changes are tracked via cryptographically signed patches, facilitating collaborative editing while preserving data integrity through a directed acyclic graph (DAG) of changes. We build upon ideas from Bitcoin [10], BitTorrent [4] and git to achieve collaborative version control in a decentralized network.

3.2.1 Immutable Data Storage with IPFS. Seed Hypermedia content can be modified by many people over time, but it is all stored as permanent data. Seed Hypermedia system uses IPFS ‘Blobs’, immutable data entries identified by a content-addressable ID (CID), to manage these changes. Three key concepts for handling these changes are Changes, Versions, and Refs.

3.2.2 Changes and Versions. A Change describes how the data of a Document may change. Each Change may contain:

Each Change is identified with an ID, a CID derived by taking a checksum of the Change data. To determine the current state of a Document with a given Change ID, you would take the following steps:

A ‘Version’ represents an immutable view of a document, explicitly referencing a set of changes. This system ensures that the document's history and state are preserved and traceable through its changes.

3.2.3 Directed Acyclic Graph (DAG) and Change Ordering. A DAG of changes defines a partial order between changes, expressed with their dependency links: a dependency change must have happened before the dependent change.

To understand how changes should be merged, it is necessary to understand the total order of Changes in a Version. The algorithm sorts first by topological order and timestamp and then by change IDs in the case of ties. The topological total order is achieved by performing a depth-first preorder traversal of the DAG of changes and sorting any siblings by the previous timestamp rules.

Timestamp Order uses the Unix timestamp expressed in microseconds carried by each change, which MUST be greater than any timestamp of any of its dependencies. In case of system clock skews, similar rules to those proposed by the Hybrid Logical Clock paper [9] are used: the lowest 16 bits of the int64 timestamp act as a logical counter, incremented (up to a threshold) until a value greater than any previously seen timestamp is obtained. If the skew exceeds the threshold, the operation is aborted, and an error is thrown, indicating a significant system clock discrepancy or a dependency change with an unrealistic future timestamp (normally prevented when applying a remote change to the local state).

Given any two changes, they are first ordered by their timestamp value, as shown in Figure 1. If the timestamps are equal (unlikely due to microsecond precision), the tie is broken by comparing the changes’ IDs lexicographically. The same author ID must avoid creating changes with identical microsecond timestamps, achievable given the microsecond precision, even for users with multiple unsynchronized devices.

To get started with Hypermedia, a peer will need to connect with a set of other peers to get acquainted with the network and find some relevant content. The most common way to join the network is to connect to a known domain over HTTPS and collect Libp2p routing information. Next, if you have a Document ID or Account ID you are interested in, you may bootstrap the Libp2p Amino DHT and use that to find peers who proclaim to have that content. Additionally, people may share routing information directly over email or messaging apps, to perform an initial connection.

Once you connect to a peer over Libp2p, you will use the built-in service discovery mechanism to upgrade your connection to the Seed Hypermedia protocol, which is built on gRPC.

When two Seed Hypermedia peers are connected, they regularly exchange information through the following process:

A peer may optionally introduce themself as an Account using a signature, which allows other peers to associate their trust information with this peer. Then, each peer will request certain content. These will include all docs and accounts to which they are currently subscribed and content they are interested in loading at the moment. In exchange, they will receive the Ref for each piece of matched content. Each peer will also broadcast a list of Document Refs and Contacts, along with metadata, including the title.

At this point, each peer has exchanged all of the relevant Refs between each-other. These Refs contain the CIDs of the latest Change Blobs for all of the content that they can exchange. Peers will use a set reconciliation algorithm to determine the full list of dependent Change CIDs they must resolve.

Last, each peer will know exactly what Blobs they need, which are synced using the BitSwap protocol from IPFS. This enables any IPFS node to resolve this data, although their current connection is probably the one best able to respond.

When a node broadcasts during the sync process, its peers may find new content that they were not already aware of. When attempting to load a Document from this list, a peer will start by initiating a sync with its current peers, which includes this Document in the request phase.

If the current peers do not have the related changes for this Ref, or if the node wants to make an effort to get a newer Ref for this Document, a peer will look up this document ID on the Amino DHT to find peers who are broadcasting it and repeat the sync process. By looping through this process with known peers and new connections, people can discover new content on the network without being inundated with spam.

It is vital to build upon known reputations to trust who you connect with. Users have the functionality to “Add a Contact” for a given Account, which performs several actions:

Connection Blobs are shared during the Sync process, allowing other nodes to see who trusts. They may contain additional metadata, including associations with domains and social networks (along with the signed proofs for these associations).

Content in the Seed Hypermedia network must be ephemeral enough to allow bad content to be flushed while ensuring permanence for content people care about.

The only way to guarantee permanence is to save it yourself. If anybody in the network chooses to archive some content, it will be retained. When some content is frequently linked, it will be archived in a viral pattern, and it becomes effectively permanent.

Nodes may privately choose to delete or block certain content from their node to prevent it from entering their system again. This is appropriate for illegal or undesirable content. A node may privately mark certain content as restricted, which will be archived permanently but will not be distributed to other peers. This may be appropriate for content that might be copyrighted, for example.

Each node may also decide what content is advertised during the broadcast phase of synchronization. By choosing what content is broadcasted, blocked, or restricted, each node can participate in moderating the content on the network and archiving important content for themselves and the world. By including the trust information from known peers and accounts, we can have high confidence that good content will spread and be preserved while bad content will be suppressed.