MosAIc, a reconfigurable reading-surface viewer for OKF bundles with LLM support

[vonargo]

Hi all! Thanks for putting OKF in the open. I'd been building the same shape independently (markdown + YAML frontmatter, git-curated, human- and agent-readable) as the memory layer of a governed AI system, so seeing it formalized was a genuinely good day.

The README frames the bundled visualize subcommand as a proof-of-concept consumer and invites better ones, so here's one.

MosAIc is a reconfigurable surface for reading an OKF bundle. Instead of a force-directed graph, it renders each concept as a panel ("tessera"), grouped by type and filterable by frontmatter.

A calm reading/working surface rather than a node-edge map. The bet is that an LLM (or a person) reasons better from a legible layout than from a graph or a wall of text.

Two things it does that lean on conversations already happening here:

• Provenance at the surface. It shows each concept's source link + freshness and visibly flags concepts that lack attribution, so "what's unsourced here?" is answerable at a glance. (Relevant to Proposal: Inline citation as default convention for coupling claims to evidence #94 / Proposal: Quality gates between layers via SOURCES convention #95 / Proposal: Elevate timestamp from optional to recommended for agent-centric staleness detection #97.)
• Point an LLM at the bundle. You can authenticate and use the loaded bundle as context (eg ask it to find the unsourced concepts, which it does).

MIT, no backend, runs entirely as a static Hugging Face Space:

• Live: https://huggingface.co/spaces/vonargo/mosAIc // https://github.com/vonargo/mosAIc
• A bit more on the thinking: https://discuss.huggingface.co/t/mosaic-a-reconfigurable-surface-where-an-llm-emits-a-layout-not-just-text/177021

Early and rough, and explicitly not trying to be the spec! MosAIc is a consumer built on it.

Happy to align frontmatter field-naming for interop, take feedback, or fold it in as a reference viewer if that's useful to anyone.