Control plane
Control plane — nous_ctl, management tree (/kernel, /eva, /memory, /workflows…)
Architecture
Two operational planes, a kernel governing the full lifecycle, and persistent memory with auditable lineage.
Architecture
Control plane separate from data plane: administration without mixing product traffic.
Data plane
Data plane — HTTP/VOICE gateway, RuntimeListener, interaction.*, runtime.*, federation
Kernel
Kernel — route_task, observers, EvaRegistry, scheduler, Avro memory + Faiss (CONTENT + STRUCTURE)
EVA
Reasoning pool (local or remote LLM) under kernel contract
EVA — Expert Virtual Agent
EVA stands for Expert Virtual Agent
EVA is NOUS’s reasoning unit: it runs steps with an LLM, catalog skills, and injected RAG, always under kernel contract. It does not administer policies, global memory, or catalogs on its own.
EVA reasons. NOUS governs.
Query-style interactions resolve structured recall without invoking EVA; tasks enter route_task and episodic memory. The pool is operated from the control plane (/eva) and can run locally or via nous.eva.socket.v1.
Persistent memory
- Episodic — compact summaries of closed tasks (Avro).
- Relational — typed graph with weights that learn from outcomes.
- Dual vector — recall by content (CONTENT) and structure/analogy (STRUCTURE).
NOUS↔NOUS federation
Multi-instance capability on the data plane: transfer experience between NOUS deployments, not via the HTTP gateway.
Export/import memory bundles (episodes + graph + indexes) with signed manifest and governed conflict resolution. federation.* ops on RuntimeListener (TCP/JSONL, PARTNER trust).
Nested Learning
Per workflow or task
1 Workflow / task
Governed instance in runtime
2 EVA reasons
Steps, tools, and catalog skills
3 Kernel observes
Policies, live confidence, recall when needed
4 Memory
Episode + graph on close
5 Audit
Lineage; zero auto-merge
On close: episode + graph update. Auditable events in the EventStore.
Consolidation between runs
When the system is idle or between operational cycles:
ImprovementsRegressionsRepeated failuresStable successes
Governed replay Decay, pattern promotion, pause meta-policy
Workflow proposal Only if governance authorizes—never auto-merge
Long-horizon memory and policies evolve without the LLM rewriting the system.
- EVA — reasoning
- NOUS — governance & observation
- Persistent memory
- Authorization & audit
Learning rhythms (Nested Learning)
| When | What |
|---|---|
| Every K inference steps | Reflective micropause (Self-RAG / FLARE) |
| During the pause | Lateral recall by structure and content |
| Scheduler ticks | Macropause: EVA health, delegation, human channel |
| Task completion | TD-update on the memory graph |
| Governed replay | Decay, promotion, patterns |
| Meta-policy | When to pause by intent |