Mooncake - Bryant McGill

# Mooncake **Entity class:** Memory infrastructure **Integration priority:** P1 **Collections:** [[collections/Consciousness Continuity|Consciousness Continuity]] · [[collections/Neurotech|Neurotech]] · [[collections/Machine Succession|Machine Succession]] **Canonical source article:** [[articles/Mind Uploading and AI — The Host is Reusable and the Person is the Delta|Mind Uploading and AI — The Host is Reusable and the Person is the Delta]] ## Definition **Mooncake** — Kimi/Moonshot KVCache-centric disaggregated serving, CPU/DRAM/SSD pool. ## Relationships  This entry is routed through [[collections/Consciousness Continuity|Consciousness Continuity]], [[collections/Neurotech|Neurotech]], and [[collections/Machine Succession|Machine Succession]]. Its source context is developed in [[articles/Mind Uploading and AI — The Host is Reusable and the Person is the Delta|Mind Uploading and AI — The Host is Reusable and the Person is the Delta]] and [[articles/Technologies for Consciousness Mapping and Transfer|Technologies for Consciousness Mapping and Transfer]]. Status-qualified source edges are preserved in the terminal Research Edges section. ### Technology and research relationships - **Mooncake** is associated with [[wiki/Kimi|Kimi]]. Moonshot AI serving platform. - **Mooncake** uses KVCache-centric scheduler. Balances throughput with SLOs. - **Mooncake** supports dynamic cache resource addition/removal. Resource composability. - The technical comparison identifies a convergence: **Mooncake** converges technically with [[wiki/Composable Infrastructure|Composable Infrastructure]]. Memory/storage resources are dynamically pooled and rebound. ### Key Relationships - **Mooncake** separates prefill and decode clusters. Disaggregated serving. - **Mooncake** is associated with KV cache across CPU DRAM and SSD. Uses underutilized cluster resources.  ## Related Work in the Corpus  - In [[articles/Mind Uploading and AI — The Host is Reusable and the Person is the Delta|Mind Uploading and AI — The Host is Reusable and the Person is the Delta]], **X. Memory became infrastructure** provides the narrative context for **Mooncake**: Underneath, the memory hierarchy has been elaborating for years in ways that map with uncomfortable precision onto what a person would need. - In [[articles/Mind Uploading and AI — The Host is Reusable and the Person is the Delta|Mind Uploading and AI — The Host is Reusable and the Person is the Delta]], **Host-infrastructure crosswalk** provides the narrative context for **Mooncake**: The active-state hierarchy is becoming infrastructure in its own right.  ## Research Edges  #### Host–residual infrastructure patch — 2026-09-10 **Collections:** [[collections/Consciousness Continuity|Consciousness Continuity]] · [[collections/Neurotech|Neurotech]] · [[collections/Machine Succession|Machine Succession]] **Canonical source article:** [[articles/Mind Uploading and AI — The Host is Reusable and the Person is the Delta|Mind Uploading and AI — The Host is Reusable and the Person is the Delta]] **Related acquisition source:** [[articles/Technologies for Consciousness Mapping and Transfer|Technologies for Consciousness Mapping and Transfer]] **Relationship source:** [[research/Mind Uploading and AI Host-Residual Ecosystem Relationship Graph - 2026-09-10|Mind Uploading and AI Host-Residual Ecosystem Relationship Graph — 2026-09-10]] #### Host-infrastructure integration Kimi/Moonshot KVCache-centric disaggregated serving, CPU/DRAM/SSD pool. #### Outgoing typed edges - **HR-0132 — `serving_platform_for` → [[wiki/Kimi|Kimi]]** — **VERIFIED EXTERNAL**; evidence `WEB_MOONCAKE`, `WEB_MOONCAKE_GITHUB`. Moonshot AI serving platform. - **HR-0133 — `separates` → `prefill and decode clusters`** — **VERIFIED EXTERNAL**; evidence `WEB_MOONCAKE`, `WEB_MOONCAKE_GITHUB`. Disaggregated serving. - **HR-0134 — `pools` → `KV cache across CPU DRAM and SSD`** — **VERIFIED EXTERNAL**; evidence `WEB_MOONCAKE`, `WEB_MOONCAKE_GITHUB`. Uses underutilized cluster resources. - **HR-0135 — `uses` → `KVCache-centric scheduler`** — **VERIFIED EXTERNAL**; evidence `WEB_MOONCAKE`. Balances throughput with SLOs. - **HR-0136 — `supports` → `dynamic cache resource addition/removal`** — **VERIFIED EXTERNAL**; evidence `WEB_MOONCAKE_GITHUB`. Resource composability. - **HR-0137 — `technical_convergence_with` → [[wiki/Composable Infrastructure|Composable Infrastructure]]** — **ANALYTIC CONVERGENCE**; evidence `WEB_MOONCAKE_GITHUB`, `WIKI_ARCHIVE`. Memory/storage resources are dynamically pooled and rebound.