System Architecture
Layer Descriptions
Signal Layer
The foundational data unit. A signal is a typed, weighted, cryptographically signed payload that replaces the concept of a “request” or “message.” Signals carry data, metadata, propagation rules, and cryptographic proofs. Learn more about Signals →Synapse Layer
Synapses are persistent, stateful connections between nodes. Unlike HTTP connections (stateless) or WebSocket connections (session-based), synapses have weight — they strengthen with use and weaken with inactivity. The synapse topology forms the neural graph of the network. Learn more about Synapses →Propagation Engine
The routing intelligence of NTL. The propagation engine determines how signals move through the synapse topology based on relevance scoring, activation thresholds, and network state. This replaces traditional routing tables and service discovery mechanisms. Learn more about Propagation →Adapter Edge
The compatibility boundary where NTL interfaces with legacy systems. Adapters translate between NTL signals and traditional protocols — HTTP, gRPC, GraphQL, WebSocket for Web2; chain transactions, DID verification, and token transfers for Web3. Learn more about Adapters →Crypto Module
A pluggable cryptographic interface that handles signal signing, verification, and encryption. The module is swappable — today it uses lattice-based post-quantum cryptography, tomorrow it can use whatever supersedes it. The transport layer never depends on a specific cryptographic scheme. Learn more about Security →Storage Interface
The bottom layer is an interface, not a database. Everything a node must remember — synapse weights and type affinity, topology knowledge, the deduplication cache, activation state, and the learning journal — is reached through the storage interface. Any backend satisfying that contract is conforming.
SQLite is the default because it needs no configuration and runs on
constrained devices — which is the edge story, not a compromise on it.
Where a graph-native backend is used, the storage graph and the synapse
topology can be the same graph: signals deposit state as they propagate, and
state changes emit signals. That is a capability of such backends rather
than an assumption of the protocol.
Learn more about storage backends →
Node Architecture
Every participant in an NTL network runs a node. A node can be a full participant or a lightweight edge node.Full Nodes
Full nodes participate in signal propagation, maintain synapse state, contribute to network topology, and persist state through a storage backend — SQLite by default, PostgreSQL where an operator already runs one.Edge Nodes
Edge nodes are lightweight participants — mobile devices, IoT sensors, constrained environments. They can emit and receive signals but delegate propagation and heavy computation to full nodes they’re synapsed with.Network Topology
NTL networks are self-organizing neural graphs. There is no central authority that defines the topology. Instead:- Nodes discover each other through bootstrap nodes (initial entry points)
- Synapses form based on signal exchange
- Frequently-used synapses strengthen, increasing propagation priority
- Inactive synapses weaken and eventually prune
- The network continuously optimizes its own topology
Design Principles
- Signal over request — Data moves as signals, not addressed requests
- Emergent over prescribed — Topology and routing emerge from usage, not configuration
- Pluggable over hardcoded — Cryptography, serialization, and adapters are all swappable modules
- Resilient over efficient — The network adapts to node failures, connectivity changes, and adversarial conditions
- Constrained-first — Designed for the hardest deployment environments, which makes it performant everywhere