AILIS Cheat Sheet (One-Pager)¶

Purpose: A crisp reference for the AI stack’s middle layers so teams can design, build, and talk about interoperable systems.

text L16 Application & Domain Logic L15 Governance, Safety & Schema L14 Session, Identity & Memory L13 Transport & Flow Semantics L12 Routing, Planning & Policy L11 Addressing & Registry L10 Tool & Function Invocation L9 Knowledge & Retrieval L8 Context Construction & Prompting L7 Inference Engine & Decoding L6 Model Parameters & Architecture L5 Tokenization & Encoders L4 Numeric & Quantization L3 ML Graph & Compilation L2 System & Driver Runtime L1 Compute Fabric L0 Facilities & Powertext

Under‑served (biggest opportunities): L11–L15.
Crowded: L1–L7, L9.

• L11 Registry: signed manifests (models, tools, indices, ensembles), capability vectors, fingerprints.
• L12 Routing: policy DSL + bandits; cascade/fan‑out/reducer; budgets & privacy classes.
• L13 Transport: idempotent runs, streaming, CANCEL/RESUME, multiplex, mid‑stream grammar negotiation.
• L14 Session: Portable Session Envelope (identity, budgets, memory refs, ensemble stack) + capability tokens.
• L15 Governance: redaction, validation/repair; schema change control; approvals & audit.

Core artifacts: manifests • capability vectors • fingerprints • routing policies • session envelopes • audit logs.

Design pattern: route → execute → evaluate → adapt (closed loop).