Live DDIL edge demo / Foundry + CASK + local mesh

DarkMesh

Altiair deploys a decentralized sensor mesh, fuses weak field evidence into one human-reviewed cue, survives node loss, and syncs the mission record back to Foundry.

DarkMesh live demo flow from sensor ingest through Foundry synchronization

Scenario

Find the useful cue without depending on one command node.

A drone-class event appears in a contested training area. One node has RFID and Wi-Fi proximity context, one has Hawkeye/vision, one has microphone context, and Foundry provides governed mission intelligence. DarkMesh turns those fragments into a policy-gated controller-zone cue for human review.

01 Mission directive

Palantir Foundry context and operator intent become a CASK deployment order with leases for each edge node.

02 Distributed evidence

Camera, microphone, RFID/provider-style proximity, and node health arrive as compact typed records instead of raw stream dependency.

03 Coordinator failover

If the active coordinator drops, the replicated ledger and gossip state let a surviving node keep the mission picture alive.

04 Commander sync

When a gateway is available, the evidence, decisions, policy gates, and coordinator terms reconcile back to Foundry.

Live mission pass

Mission directive to edge coordination.

Detecting source Waiting for CASK state
DM

DARPA field prototype / Altiair local mesh

DarkMesh Mission Copilot

--:--:-- Local clock -- Mission readiness -- Last fusion

Mission

-

Status

-

Confidence

-

Latest fused event

-
Friendly Node
Degraded Node
Mesh Link
Camera FOV
RF Bearing
Fused Detection
Objective Area
Track Trail
Mission mesh map Nodes, mesh links, sensor fields of view, objective area, and fused event estimate.

Cue chain

Your next action

-

Evidence summary

-

Live at the bench

Running right now on a Jetson Nano, three Raspberry Pis, and an iPad chest computer.

A five-node fleet on the team bench, plugged in and running right now: Jetson Orin Nano altiair-orin, Pi 5 altiair-hub, two Pi 4B sensor peers, and an iPad chest harness mirroring the operator console. Same Altiair-LAN, same node API, same Gemma 4 e2b on-device inference. The dropout demo is triggered over SSH — judges never watch us unplug a Pi. The mesh stays powered through the whole pitch.

  • altiair-orin Jetson Orin Nano · LAN host · accelerated inference · gateway
  • altiair-hub Pi 5 · operator display + USB camera · primary gateway
  • altiair-node-a / -b Pi 4B · sensors · RFID · replicated ledger
  • iPad chest harness Mirrors the operator console · wearable EagleEye readout
  • Local Wi-Fi mesh No cloud, no internet path
  • gemma4:e2b Ollama-served on-device
  • Survivable Replicated ledger + Raft-style coordinator
Edge fleet at the team bench: Jetson Orin Nano, Raspberry Pi 5 hub, two Pi 4B sensor peers, and iPad chest-worn operator console, plugged in and running on local Altiair-LAN over Wi-Fi
Live: hardware plugged in and running at the bench right now. Node-dropout demo is triggered over SSH — no Pi gets unplugged.

Working code path

The website walks through the running system.

The page reads the same node API that the Pi and Jetson nodes expose. During the live pass it shows each code path, endpoint poll, ledger update, coordinator term, local model result, node-dropout state, and Foundry writeback boundary as the mission advances.

execution feed DarkMesh runtime

System architecture

Robust, safe, secure edge coordination.

Foundry and CASK shape the mission data. DarkMesh keeps the local runtime useful when the uplink or one node fails, while policy gates, signed records, encrypted payloads, and memory-safe durable components keep the demo on a production path.

Team

Built by Altiair.

Four engineers, twenty-four hours, one DDIL mission stack.

Team Altiair at NatSec Hackathon 2026, Shack15 SF
NatSec Hackathon 2026 · Shack15, San Francisco

Source · github.com/digitalnomd/altiair   ·   Site · github.com/benikigai/darkmesh