A distributed cognitive architecture built across heterogeneous hardware: Jetson edge devices, a 4070 "brainstem" node, a NAS-backed long-term memory substrate, and a 4090 "cortex" reasoning engine.
Nexus models concepts from biological cognition (perception, filtering, memory consolidation, hierarchical reasoning) as a modular, containerized system. It is not a single model; it is an ecosystem of cooperating processes.
- Distributed synthetic cognition across commodity GPUs and edge devices
- Persistent identity structures over long time horizons
- Long-term memory substrates for LLMs (semantic, episodic, schema-driven)
- Multi-node orchestration with low-latency cross-device inference
- Memory filtering, abstraction, and decay
- Safe agent behavior via layered gating and instinctual rules
Biological intelligence is distributed. Nerves preprocess. The brainstem filters. Memory consolidates. The cortex reasons. Nexus recreates that division of labor in software.
- Jetson nodes (peripheral nervous system). Capture raw video, audio, telemetry. Perform early filtering and compression. Push signals to the brainstem.
- 4070 node (brainstem). Validates incoming signals, generates embeddings, applies instinctual rules, buffers short-term memory, and decides what is important enough to consolidate. Since Sprint 5 it also serves as the Nexus Hub, the family-of-models orchestration layer:
GET /family(roster + presence + queue depths),GET /members/{id}(spec summary, presence, model info),POST /members/{id}/chat(200for a live reply,202+msg_idwhen the message is queued,503 member_loading+Retry-Afterwhen the member is waking up),POST /members/{id}/presence,GET /members/{id}/inbox/{msg_id}(poll a queued message's reply), andPOST /members/{id}/memory/promote(share a private memory to the household). Seedocs/architecture_v2_family_of_models.mdfor the full design. - NAS (long-term memory). Vector database for semantic memory, time-ordered episodic logs, knowledge graph structures, decay and deduplication. Synthetic hippocampus.
- 4090 node (cortex). Hosts large-scale LLM reasoning, executes high-level planning, integrates episodic and semantic recall, and orchestrates downstream agents.
- Model manager (4090 node, Sprint 5).
nodes/model_manager_4090/— owns weights placement and thellama-serverlifecycle for whichever family member is resident. Tiers weights hot/warm/cold (NVMe Gen4 / NVMe Gen2 / HDD) viaensure_hot()staged, checksum-verified copies, since llama.cpp does not tier for us; loads/unloadsllama-serverper member and reports presence (waking/awake/etc.) back to the hub. - Consolidation engine (sleep node). Re-embeds old memories, clusters and abstracts concepts, summarizes logs into narratives, enforces schema consistency, runs "synthetic dreams".
Nexus is developed alongside three research works covering:
- Synthetic continuity and identity persistence
- Biologically-inspired distributed cognition
- Containerized intelligence and propagation patterns
Drafts live in papers/.
Phase 1: core bring-up. Inter-node message bus, Jetson to 4070 to NAS pathway, vLLM / llama.cpp on the 4090, memory schemas and storage logic.
Phase 2: multi-node coordination. Real-time sensory processing, brainstem instinct engine, semantic and episodic write pipelines, cortex RAG integration.
Phase 3: autonomous consolidation. Replay cycles, schema enforcement, summarization, daily sleep routines.
Phase 4: safety and alignment layer. Rate limits, instinctual rules, safe-operation constraints, preference stability.
Nexus is an evolving research-grade architecture. Contributions, ideas, and code reviews are welcome as the system matures.
- Author: Andrew "Drew" Mattick
- GitHub: Ginkobaloba
- Marketing site: projectnexuscode.org
See LICENSE.