A research map for the AI ecosystem.

AILIS (AI Layer Interface Specification) is a Dollhouse Research white paper exploring a 16+ layer vocabulary for AI systems, from physical compute up through orchestration, governance, protocols, products, and domain behavior.

Read the Primer Open the Cheat Sheet Join the Discussion

Layer Model

The first thing to see is the stack.

The draft groups the AI ecosystem into practical zones. It is asking whether this kind of map could make architecture discussions clearer, especially where infrastructure, model behavior, protocols, orchestration, governance, and products blur together.

AILIS is not trying to enforce a compliance model, and it is expected to change as people test it against real systems.

Research Context

AILIS is one research project from Dollhouse Research.

This site is focused on AILIS: a public white-paper effort for describing AI system layers. DollhouseMCP and MCP-AQL are related public projects from the same research home, shown here as portfolio context rather than dependencies of AILIS.

Research Home

Dollhouse Research

An applied AI systems research home for public papers, protocols, infrastructure work, and experiments as they become ready to share.

Open Dollhouse Research

Current project

AILIS

White paper and vocabulary for understanding AI system layers.

Read the Primer

Peer project

DollhouseMCP

Distributable MCP server with broad capability and safety controls.

Open DollhouseMCP

Peer project

MCP-AQL

Protocol layer on top of MCP for semantic routing and introspection.

Open MCP-AQL

Related Example

Semantic routing as a layer-boundary example.

MCP-AQL is useful context for AILIS because it shows how protocol semantics, safety classification, runtime discovery, and token economics can be separated from application behavior.

MCP-AQL concern	AILIS lens	Why it matters
CRUDE endpoints	L10-L13 tool, routing, and transport semantics	Operation intent is explicit before a client acts.
Runtime introspection	L11 registry and capability discovery	Clients can discover valid operations on demand.
Permission classification	L12-L15 policy, safety, and governance	Read, delete, and execute paths can receive different controls.
Token reduction	L8-L10 context construction and tool exposure	Large tool menus do not have to be loaded up front.

Quick Path

Start here if you are new to AILIS.

10 minutes

Read the Primer

Get the full layer model, motivation, and open questions.

Open Primer

2 minutes

Use the Cheat Sheet

Scan the layer definitions and cross-cutting control planes.

Open Cheat Sheet

Contribute

Share critique

Challenge the model, submit a use case, or propose a different framing.

Review feedback areas

Build locally

Preview the site

Use MkDocs to run the documentation site and validate changes.

Open website workflow

Open Questions

What we are asking the community.

Resonance
Does this framing make AI architecture easier to discuss, or does it feel forced?
Boundaries
Which layers are missing, overloaded, or miscategorized?
Alternatives
Are there better ways to describe these interfaces and control planes?
Evidence
Which real-world systems should be mapped first to test whether AILIS holds up?