| status | EMERGING | |||
|---|---|---|---|---|
| last-verified | 2026-04-28 | |||
| evidence-tier | C | |||
| applies-to-signals |
|
|||
| revalidate-by | 2026-10-28 |
Evidence Tier: C — federation is hand-rolled today; the OpenBrain compilation-agent path is Tier D (not shipped).
Per-archetype recommendation for managing multiple knowledge-bearing repos as a portfolio — preserving per-repo evidence and provenance while still surfacing cross-repo connections.
Per-repo (LLM Wiki + Graphify) + a thin federation index. Federation today is hand-rolled. OpenBrain's compilation agent is roadmap, not shipping.
| Layer | Owner | Why |
|---|---|---|
| Per-repo | Karpathy wiki + Graphify (as in archetype A) | Preserves per-repo evidence tier and provenance |
| Federation | Small ~/portfolio/INDEX.md + nightly cron merging per-repo index.md files |
Axis 4 — federation, not consolidation |
| Cross-repo topology | Second graphify run over all repos → cross-repo communities tagged INFERRED | Axis 8 — separates within-repo (EXTRACTED) from cross-repo (INFERRED) |
Driving axes: 4 (cross-tool/cross-repo), 2 (augments per repo), 8 (provenance discipline survives federation). Tier: C; OpenBrain compilation-agent path is Tier D (not shipped).
| Hybrid | Optimizes | Pick when |
|---|---|---|
| Per-repo + OpenBrain federation (when compilation agent ships) | Live cross-repo recall | FSL-1.1-MIT terms acceptable; want concurrent multi-tool reads |
| Per-repo + claude-context only on code-heavy repos | Semantic recall in code without forcing every repo into vectors | Some repos are code, some are prose |
| Single mega-repo with tagged subfolders | Drops federation complexity | Repos are small and you're really managing one knowledge body |
- Single-DB consolidation across all repos (OpenBrain Postgres, claude-context Milvus): flattens per-repo evidence tiers; one repo's schema decisions infect another's. A single Milvus collection with vectors from prose + AST chunks produces nonsense neighbors.
- Shared cross-repo graph as the only graph: communities bleed across project boundaries. The cross-repo graph must be a secondary INFERRED-only view.
- Each repo independently passes archetype-A adoption first.
- Write a 50-line script concatenating per-repo
index.mdfiles into a portfolio INDEX. - Run
graphify ~/portfolioover all repos to see what cross-repo edges emerge. Stop if edges are mostly noise. - Re-evaluate OpenBrain only after the compilation agent ships and you've reproduced its claims.
All six met when run as per-repo + thin federation. OpenBrain path defers local-first to self-hosted Postgres setup; FSL terms must be accepted.
Inherits source rubric and tier methodology from memory-systems-recommendation-methodology.md.
This variant carries no new external sources with independent URLs beyond what the methodology doc and Archetype A already cite. The per-repo layer inherits Karpathy (Tier B paradigm) and graphify (Tier C, safishamsi/graphify) from Archetype A. The OpenBrain compilation-agent path is explicitly labeled Tier D (not shipped) in the doc body.
memory-systems-archetype-recommendations.md— index across all 7 archetypes + cross-cutting sectionsmemory-systems-recommendation-methodology.md— framing and self-critiquefederated-query-architecture.md— federation patterns relevant herecross-project-synchronization.md— cross-project sync patterns
analysis/memory-systems-archetype-recommendations.md[EXTRACTED (1.00) ×4] — referencesINDEX.md[EXTRACTED (1.00)] — referencesanalysis/memory-systems-archetype-b-code-monorepo.md[EXTRACTED (1.00) ×2] — referencesresearch/memory-systems-project-archetypes.md[EXTRACTED (1.00)] — referencesresearch/memory-systems-tools-inventory.md[EXTRACTED (1.00) ×2] — referencesanalysis/memory-systems-archetype-f-session-archive.md[INFERRED (0.75)] — semantically_similar_toanalysis/memory-systems-archetype-e-work-state-tracker.md[INFERRED (0.70)] — semantically_similar_to