L11 - Addressing & Registry¶
L11 covers discovery, naming, manifests, capability metadata, signed artifacts, provenance, versioning, fingerprints, and registries for models, tools, datasets, prompts, adapters, indices, and ensembles.
L11Addressing & Registry
- Discovery
- Manifests
- Signatures
- Capability metadata
What belongs here¶
L11 asks how a system knows what exists, what version it is, what capabilities it claims, what evidence supports those claims, and whether the artifact can be trusted. AILIS currently treats this as one of the more under-served middle layers.
Representative projects¶
| Project | Why it might fit | Adjacent layers |
|---|---|---|
| Hugging Face Hub | Registry and collaboration platform for models, datasets, and spaces. | L6 models, L11 registry |
| Docker Hub and OCI registries | Image naming, versioning, and distribution patterns that may inspire AI artifact registries. | L2 runtime, L11 registry |
| MLflow Model Registry | Model versioning, lifecycle, and registry workflow. | L6 models, L15 governance |
| Sigstore | Signing and provenance infrastructure relevant to artifact trust. | L11 addressing, L15 governance |
| OpenAPI descriptions | Machine-readable API descriptions that can be registered and discovered. | L10 tools, L11 registry |
| Model Context Protocol server ecosystem | MCP servers raise registry and capability-discovery questions for AI tools. | L10 tools, L13 transport |
| Dollhouse Collection | Public metadata and activation surface where catalog entries, artifact descriptions, and discovery cues act like a lightweight registry. | L9 retrieval, L16 product |
| MCP-AQL spec and generator tooling | Introspection, manifest generation, and capability-description work that makes protocol artifacts easier to name, inspect, and compare. | L11 registry, L15 schema |
Boundary questions¶
- What is the minimum useful manifest for a model, tool, index, prompt, or ensemble?
- Should capability vectors be registered claims, measured evidence, or both?
- How should registries represent incompatible versions, deprecations, and security advisories?
Signals to watch¶
- Signed AI artifacts becoming a default expectation.
- Capability metadata moving from marketing claims into machine-readable evidence.
- Tool and model discovery becoming provider-neutral rather than application-specific.