Content mounts and federation

The content-mount pattern combines two authoring modes in a single wiki instance: curated editorial articles committed directly to the knowledge repository, and content mounted from a separate local directory. A reader browsing the wiki cannot tell which mode produced any given article. Both render into the same category pages, the same search index, the same navigation, with no visible seam. Local directory mounting is live today; a further extension — mounting content fetched automatically from a remote git repository rather than an already-local path — remains a design direction, not a shipped capability.

The hybrid model

A deployed wiki instance is configured by a knowledge.toml manifest at the content directory root. The manifest declares one or more mounts, each pointing to a local directory, alongside any articles committed directly to the repository. Exactly one mount may hold the primary role — the editable article set; any additional mounts are read-only. Both render into the same URL namespace.

Direct commits into the primary mount are the correct choice for content that is editorial in nature — platform reference articles, design-system notes, governance documents — where the knowledge repository is the canonical home and editing flows through the standard staging-tier commit path. A read-only mount is the correct choice for content whose canonical home is a different repository entirely — the wiki surfaces it without taking on editing responsibility for it.

Why it matters: a wiki instance can surface content it does not own without ever becoming responsible for editing it — the canonical home stays wherever the content actually lives.

How mounts work today

A mount entry in knowledge.toml declares a local filesystem path, a role (primary or read-only), and a blueprint_set — a list of content-type names the mount is expected to contain. The engine walks each declared mount's directory at startup and builds its content index from what it finds there; a mount's directory must already exist on disk when the engine starts.

What this does not yet do: fetch content from a remote repository on the wiki's own initiative. A mount today is a pointer to an already-present local directory, not a git remote the engine clones or pulls on a schedule. Getting content from a separate repository into a mount's directory is a step performed outside the wiki engine — by whatever process keeps that directory populated — not something the engine does for itself.

Why it matters: an operator evaluating this pattern today should expect to keep a mount's directory populated themselves — the engine reads what is already there, it does not go fetch anything on its own yet.

Blueprints

The blueprint_set field lets a mount declare which content types it is expected to hold — TOPIC and GUIDE are the two names in current use. This is a declared expectation carried in configuration; it does not yet drive content-type-specific validation or routing behavior beyond that declaration.

Why it matters: a reader inspecting a mount's configuration can tell what kind of content to expect there without opening a single file — the declaration is the documentation.

Per-instance isolation

Each wiki instance reads only the mounts declared in its own knowledge.toml. Two instances could in principle point their mounts at directories sourced from the same upstream content, and still present entirely different article sets depending on their own mount configuration — isolation is a property of each instance's own configuration, not something coordinated centrally.

Why it matters: misconfiguring one wiki instance's mounts can never leak content into, or pull content out of, a second instance — each instance's article set is entirely its own configuration's doing.

What a remote-fetch extension would add

Extending local-path mounting to remote-repository fetching — the wiki engine cloning and periodically refreshing a git remote into a mount's directory itself, rather than relying on an external process to keep that directory current — is a real design direction for this pattern, not yet built. The natural extension alongside it is per-article provenance metadata (which source repository and path an article came from) and edit-URL routing (sending an editor back to the source repository's own edit path rather than accepting a write locally) — neither of which exists in the mount configuration today.

Why it matters: naming this as a planned direction rather than a shipped feature means a reader is never misled into thinking the engine already manages remote repositories on its own.

Relationship to source-of-truth inversion

This pattern extends Source-of-truth inversion from a single-repository topology toward a multi-repository one. The invariant is the same: git is canonical, the running binary is a view. What a mount adds is that "the canonical repository" is no longer necessarily the wiki's own repository — a mounted directory's real canonical home can be wherever the process populating it draws from.

Why it matters: a reader who already understands source-of-truth inversion does not need a second mental model for mounts — it is the same canonical/view invariant, just with more than one canonical repository in play.

See also

• MediaKit knowledge application — the engine that implements this pattern
• Source-of-truth inversion — the foundational design choice this pattern extends
• Knowledge wiki leapfrog architecture — the broader IA and navigation philosophy