Hypermedia blobs — Onyx on real data
The Hypermedia Network stores its data as DAG-CBOR blobs in IPFS. There are six signed blob types — Change, Ref, Profile, Comment, Capability, Contact — and they are related: every one embeds the same signed envelope. This is a real, production schemafication built entirely from the Onyx features in this repo, and it lives under its own authority, hm://seed.hyper.media/* (local files hypermedia-*.json).
The shared base — extension in action
Every blob embeds a base envelope, Signed blob:
field | type | meaning |
|---|---|---|
| string | the blob discriminator ("Change", "Ref", …) |
|
| the signer's public key |
|
| signature over the blob |
|
| Unix-millisecond time |
Each concrete type extends it (Onyx extension — the schema language), inheriting those four fields and overriding type with a single-value enum:
hypermedia-change — an append-only document change, linked into a causal DAG by deps; carries a change-body of ops.
hypermedia-ref — a signed pointer from a space/path to the current head Changes.
hypermedia-profile — an account's name / avatar / description (or an alias).
hypermedia-comment — a threaded comment; body is a tree of comment-blocks.
hypermedia-capability — a delegation of a role (WRITER / AGENT) to a key.
hypermedia-contact — one account's named reference to another.
Open hypermedia-change in the schema explorer: signer/sig/ts show as inherited, the rest as added. That's the "block types are related" relationship, made visible — and hypermedia-blob's Dependents list is exactly those six.
Define your own signed blob type
The envelope is not reserved for the six built-in types. Any schema that extends Signed blob and pins a type tag is a signed blob type the app knows how to create: in the schema editor, tick Signed blob type, set the tag (say Vote), add your fields, and publish the type as a schema-definition page. Its Create button then opens the signing form — you fill in only your fields; signer, ts, and sig are added at signing time with the selected account's key, and the blob is published with the same convention the daemon verifies (sign the canonical CBOR with the signature zeroed). The result is a first-class, verifiable blob on the network — with a type the built-in indexer ignores but any app that resolves your schema can trust and render. Browse any schema, built-in or yours, at /hm/schema/<cid>.
The union — one of six
hypermedia-any-blob is the discriminated union of all six, tagged on type — so "any Hypermedia blob" is a first-class type you can validate against.
Nested structure
A Change's body is a list of ops — themselves a union (SetAttributes / MoveBlocks / ReplaceBlock / DeleteBlocks / SetKey). Content is modeled by block and annotation, both open maps (known fields + arbitrary inline attributes). Document metadata is an open struct of known keys (name, summary, icon, cover, layout, …) plus extras.
Block types — a strict core anyone can extend
Document content is made of blocks. We want two things that pull in opposite directions: strict, concrete types (so implementations can dispatch on block.type with type-safe, per-type handlers) and openness (so a newer client's block type doesn't make an older client reject the whole document). These can't both live in a single validation pass — an open fallback always swallows a malformed known block — so the model provides layers, and you pick per workflow:
workflow | needs | use |
|---|---|---|
rendering / dispatch | strict per-type shapes + graceful fallback | concrete types + |
authoring / editing | strict validation |
|
sync / storage (forward-compat) | never reject unknown |
|
codegen | the enumerable set |
|
The fifteen concrete blocks — hypermedia-block-paragraph, hypermedia-block-heading, hypermedia-block-code, hypermedia-block-math, hypermedia-block-image, hypermedia-block-video, hypermedia-block-file, hypermedia-block-button, hypermedia-block-embed, hypermedia-block-web-embed, hypermedia-block-nostr, hypermedia-block-table, hypermedia-block-table-row, hypermedia-block-table-column, hypermedia-block-query — each extends hypermedia-block-base, closed, with a type enum and typed attributes.
hypermedia-block-core — the core union we define (the fifteen). Strict: rejects anything else.
hypermedia-block — the open block: id + type + arbitrary fields (via onyx-any). The forward-compatible wire type — a block type this client has no schema for (future or third-party) is still a valid Block, so a document is never rejected over it. This is not "your custom block type" (that's just extension + union, below); it's the open fallback for the unknown.
Adding a block type
To add a block type, do exactly what the core blocks do — extend hypermedia-block-base — then union it with the core. No new machinery:
// example-app-block: the core, PLUS this app's custom Poll block
{ "anyOf": [ { "ref": "hm://seed.hyper.media/block-core" },
{ "ref": "hm://example.com/poll-block" } ] }See example-poll-block (a custom block extending the same base) and example-app-block. That union is strict for its app — it accepts core blocks and Polls but rejects a block type it doesn't know — while the wire's hypermedia-block stays open.
Change is generic over its block type
To make Change itself strict over an app's block set — not just the wire block — hypermedia-change is a Change<Block>: the Block parameter threads through change → change-body → op → op-replace-block (each level passes it down with args), defaulting to the extensible hypermedia-block. An app instantiates it — example-myapp-change = Change<example-app-block> — and now a ReplaceBlock op carrying a block type the app doesn't know is rejected four levels deep ($.body.ops[0].block), while the default Change still accepts anything. This is real generic abstraction (params / var / args); see the schema language.
CBOR value shapes
The wire types map onto Onyx primitives, wrapped as self-explanatory aliases:
Hypermedia | CBOR | Onyx |
|---|---|---|
| byte string |
|
| CBOR tag-42 link |
|
| int64 (Unix ms) |
|
Every one of these schemas is validated in validate.mjs — as a well-formed schema, and against real blob-shaped data (a Ref, a Capability, a Change with ops, the union, and metadata), with negative cases for wrong type tags, missing required fields, and unknown keys.
Seed API read models
Beyond the signed blobs, the seed-* schemas type the derived data the Seed daemon computes for clients — not signed network data, but the read models the apps consume: seed-resource (the union of every state a fetched resource can be in: document, comment, redirect, not-found, tombstone, error), seed-document and seed-comment (the API payload forms, with resolved versions, authors and timestamps), seed-citation, seed-interaction-summary, seed-search-results, seed-site-member, seed-contact-record, and seed-discovery-status, built on seed-id (the parsed form of an hm:// identifier). Typing these closes the loop: the same ontology describes what is signed on the wire AND what the API serves back.
Do you like what you are reading? Subscribe to receive updates.
Unsubscribe anytime