Architecture

Architecture

Generated with Grok Build: Grok 4.5 · xAI Imagine (/imagine)

Layout

thalamic-relay/
├── src/
│   ├── main.rs      # CLI, lockfile, main loop, UDP, SNN step
│   ├── lib.rs       # re-exports cpu + gpu modules
│   ├── gpu.rs       # HardwareBridge, GpuTelemetry, safety, emergency brake
│   └── cpu.rs       # init_telemetry, RelayMetrics, metrics collector
├── Cargo.toml       # bin thalamic-relay; neuromod, tokio, nvml-wrapper, …
└── .github/workflows/ci.yml

Standalone crate (no workspace members). Former FPGA/silicon-bridge coupling removed — modularity over in-tree hardware export.

Main loop (simplified)

1. Parse CLI; acquire /tmp/thalamic_relay.lock before binding ports
2. init_telemetry → tracing + Prometheus listener
3. Build SpikingNetwork::with_dimensions(num_lif, num_izh, num_channels)
4. Bind UDP; each step:
   • Read telemetry (read_telemetry_force)
   • Handle emergency brake apply/release tasks
   • Every 10 steps: check_safety
   • process_udp_messages (stimuli / reward / GetNeuroState)
   • network.step(&stimuli, &modulators)
   • Update RelayMetrics / dashboard
   • Sleep step_interval_ms

Runtime stack

Piece	Crate / type
Async	tokio
SNN	neuromod::{SpikingNetwork, NeuroModulators}
GPU	nvml-wrapper + nvidia-smi fallbacks
Metrics	metrics + metrics-exporter-prometheus
Logging	tracing / tracing-subscriber
CLI	clap derive + env

Related

• UDP IPC
• GPU Telemetry
• SNN Integration

---

Last updated: July 29, 2026 Updated by: Grok Build: Grok 4.5 Package tip reference: e09cd4a (main) Devin DeepWiki: commit e09cd4a · thalamic-relay