Knowledge

Every stack can have a knowledge base

Every stack on stacknet can attach knowledge sources which are external, verifiable reference collections that the stack’s AI pipelines can search and cite. A knowledge source can be pointed or ran yourself. It ultimately is a URL that uses the pedia protocol, a small six-endpoint spec that any provider can implement resulting in exposing of the knowledge surface to the stack.

stack.metadata.knowledgeSources: [
  {
    name: "stackipedia",
    url:  "https://pedia.stacknet.org/api/v1/pedia",
    protocol: "pedia",
    addedAt: 1776231400000,
    lastVerifiedAt: 1776317800000,
    merkleRoot: "a3f2c1..."      // SHA-256 hex of the provider's manifest root
  },
  ...
]

The full reference collection (articles, sections, references, categories) lives in the provider. Stacknet records the handshake and authenticates the data integrity. When a stack queries its knowledge, the call goes to the provider; stacknet never holds or caches the content itself. The data should be data that can be ultimately public.

The pedia protocol

Six endpoints, documented and stable:

A provider that implements these six endpoints is a valid knowledge source. Stacknet doesn’t care about the storage engine (Geoffipedia uses SQLite), the ingestion pipeline, or the content format. The protocol is the contract.

Merkle verification: trust, but verify

Every pedia provider publishes a merkle root over its article collection. When a stack adds a source, the network records the current root stack.metadata.knowledgeSources[i].merkleRoot, and stamps lastVerifiedAt.

Per-stack isolation

Knowledge sources are scoped to a single stack, not the network. Two stacks owned by the same user can subscribe to the same provider with independent configurations. A public knowledge source subscribed by stack A is invisible to stack B. There is no global knowledge registry today: the source URL is the registry.

This falls out naturally from the storage model.