Neuralware · neuralware
Owner directive (2026-06-12). "Neuralware is a plug-an-play Cogni with due source and qasi-kernel resources. So it is of extreme importance the concepts,: tools, skils, whatever… are well set and ready for the Cognitive Loop when some Looper or Datan chose a Neuralware." [src: wiki/development/prompts-archive/by-topic/knowledge-management-and-concept-modeling.md:184-185] (Definition is owner-ruled; the catalog COUNT is not — hence status: draft, see Open question.)
Content
A Neuralware is a plug-and-play cogni — "a capability bundled with its due source and quasi-kernel resources — that a Looper or Datan equips into the Cognitive Loop. Neuralwares are the third stage of the Cognitive Loop: the Neural Engine runs the fixed sequence Cognis → Kernels → Neuralwares on every loop." [src: wiki/reference/neuralware-catalog.md:3]
The contract — every neuralware carries: capability (what the datan can now do), source ("the authoritative origin (docs/code/spec) — due source, never guessed"), quasi-kernel ("the runtime resources the capability needs… a quasi-kernel because it plugs into the host Cogni's real kernel rather than owning one"), and compatibleNetworks [src: wiki/reference/neuralware-catalog.md:9-14]. Equipping is enforced in code — equipNeuralware throws if the Datan's network is incompatible [src: wiki/reference/neuralware-catalog.md:16]; since 2026-06-12, "Every capability is made available to every Datan (available_to: all)" [src: wiki/reference/neuralware-catalog.md:3].
Gate discipline: a neuralware's source + quasi_kernel must resolve at gate-time; a half-formed neuralware fails the cognitive-loop gate and stalls the loop [src: wiki/reference/neuralware-catalog.md:63]. When equipped, it is "injected as an additional Cogni in the Cognitive Loop traversal" for the rest of that loop only [src: wiki/reference/neuralware-catalog.md:63-72]. If a Datan cannot deliver, it "should go to the academy and bring a new neuralware… to try again" [src: wiki/development/prompts-archive/by-topic/knowledge-management-and-concept-modeling.md:612]. Curated loadouts of neuralwares are gear-sets; the full per-item catalog is neuralware-index.
Variants
| # | Definition (verbatim) | Source | Ruling | Status |
|---|---|---|---|---|
| A | "count": 37 — "37 neuralwares (20 original + 3 web-capability + 5 browser-platform + 9 ontology/tooling additions)" | [src: wiki/voidweb/neuralwares.json:3] [src: wiki/reference/neuralware-catalog.md:5] | No ruling found — no D-entry or C-entry fixes the canonical count | draft |
| B | 28 entries — the code registry before RUN-013 #11 (20 original + 3 + 5, without the 9 doc-only ontology/tooling items); still shipped in concept-validation/lib/src/lib/sapiens3/neuralware-registry.ts (28 nw- entries, verified 2026-07-10) | [src: concept-validation/lib/src/lib/sapiens3/neuralware-registry.ts:3] | No ruling found | draft |
| C | HOS-Webdoc registry now carries 37 — "Canon reconciliation (RUN-013 #11) — 9 neuralwares declared in wiki/voidweb/neuralwares.json (37 total) but previously doc-only" annexed into NEURALWARE_REGISTRY (37 nw- entries, verified 2026-07-10) | [src: HOS-Webdoc/web/src/lib/sapiens3/neuralware-registry.ts:51-62] | Build-forward annex, not an owner ruling on the canonical count | draft |
Internal catalog drift, noted for the ruling: the catalog header says "20 original" [src: wiki/reference/neuralware-catalog.md:5] while its own section heading says "The 19 originals" [src: wiki/reference/neuralware-catalog.md:49] (the table lists 18 network-scoped + Loop Runner + Spark Emitter under "Universal").
Open question
To the owner: what is the canonical neuralware count and registry of record? Candidates: (a) 37 per wiki/voidweb/neuralwares.json (count field) and the HOS-Webdoc registry post-RUN-013; (b) 28 per the pre-annex code registry still shipped in concept-validation. Also: is neuralwares.json or neuralware-registry.ts the single source of truth going forward, and is the "original" set 19 or 20? No D/C entry rules this; per SOP the card stays draft — see sop-002-conflict-resolution.
Composition
- uses: cogni (a neuralware IS a plug-and-play one), kernel (quasi-kernel plugs into the host Cogni's kernel), cognitive-loop (third stage; gate-checked)
- used-by: datan / looper (equip them), neural-engine (pulls each Datan/Neuralware model and runs it), gear-sets (curated loadouts), neuralware-index (the catalog card), squad (dynamic squads swap neuralwares per particle)
See also
- neuralware-index
- gear-sets
- [Upstream wiki page](../../../wiki/reference/neuralware-catalog.md)