A multi-agent orchestration system that takes content and results in, and responds by curating a totally agent-controlled memory: the brain. Built local-first by a single operator; part of Project Ava.
Honest status (2026-06-06): this repo was
cheatSheets(a personal learning system) through v7.8.1. An 18-month audit found elaborate architecture with no closed loops, and the project was re-founded as Cortex with the learning system demoted to one consumer surface. The full ontology, status-tagged so aspiration cannot masquerade as fact, lives inSPEC.md. The SPEC itself is an unvalidated hypothesis until Loop 1 closes (PLAN.md).
flowchart TB
S["SURFACES: agent-first UI, learning companion, tutor (PLANNED/STALLED)"]
O["ORCHESTRATION: Claude brain + local-model workers + OpenClaw host (PARTIAL)"]
M["MEMORY: corpus mirror + SQLite/FTS5 + vectors + continuity + consolidation (RUNS/STALLED)"]
I["INGESTION: crawl -> triage -> curate -> polish -> embed (RUNS, in sibling hub)"]
I --> M
M <--> O
O --> S
S -->|results, usage, feedback| I
- Ingestion
RUNS: a 6-stage human-gated pipeline (12 sources, ~680K chunks, hybrid vector+BM25+symbol retrieval) currently lives in the sibling Ava hub; porting it here is Loop 1 work - Memory
RUNS substrate / STALLED curation: markdown corpus mirror (human-auditable, Obsidian-friendly), SQLite + FTS5, ChromaDB vectors, and a continuity ledger. The consolidation layer (Hippocampus: sleep-phase promotion of memories) is built but starved - Orchestration
PARTIAL: premium model as the judging brain, local models as bulk workers behind MCP tools; zero new frameworks by decision - Surfaces
PLANNED: an agent-first UI + backend; the former cheatSheets learning companion revives when content exists
Every component in every document carries a status tag (RUNS / STALLED / PLANNED / SPECULATIVE). Present tense is permitted only for RUNS. That discipline is a survivor's rule: see the post-mortem notes in plans/archive/cortex-rebrand/.
cortex (repo: cheatSheets)/
├── SPEC.md # Ontology + identity (single source of truth)
├── CLAUDE.md # Session rules + dev orchestration model
├── PLAN.md # Index of active work
├── plans/ # Active plans (currently: Loop 1) + archive/
├── exploration/ # Ideas, research, resource landscape, idea-mines
├── architecture/ # Consolidated/completed work + retired docs (with receipts)
├── DECISIONS.md # Curated architectural decision record
├── vault/ # Obsidian vault (learning-companion surface, STALLED)
├── knowledge-agents/ # Agent identity packs (learning surface, not yet exercised)
├── archive/ # Scoped: ELEGOO Mega 2560 reference kit
└── .ava/ # Curriculum engine product files (runtime gitignored)
SPEC.md: what Cortex is: memory model, layer ontology, entity catalog ordered by what actually exists, Loop 1 commitmentexploration/resource-landscape.md: every technology considered, with ADOPT/TRIAL/DEFER/REJECT verdicts and named triggersPLAN.md: the active work indexDECISIONS.md: why things are the way they are
Released under the MIT License.