JetsonClaw1 Vessel is a Rust-based agent vessel runtime for NVIDIA Jetson edge devices, providing the foundational identity, capability declaration, and communication primitives for a fleet-managed autonomous agent node.
Edge AI devices like the NVIDIA Jetson series are increasingly deployed in fleet configurations for robotics, environmental monitoring, and industrial automation. Managing identity, capabilities, and inter-agent communication across heterogeneous edge hardware requires a structured vessel abstraction. JetsonClaw1 Vessel implements the GIT-AGENT-STANDARD protocol, enabling a single Jetson node to self-describe its capabilities, maintain a duty diary, exchange knowledge with fleet peers, and participate in the SuperInstance task dispatch system. Without a vessel layer, edge agents are isolated silos; with it, they become addressable, inspectable, and coordinated participants in a larger computational ecology.
The vessel architecture follows a declarative identity model. On startup, the agent reads IDENTITY.md to establish its persistent self-description, then registers its CAPABILITY.toml with the fleet orchestrator. The capability file uses TOML because it is human-readable, merge-friendly, and maps cleanly to Rust's serde deserialization.
Communication follows the bottle protocol — immutable messages passed between agents. Each vessel maintains:
for-fleet/— outbound messages queued for fleet distributionfrom-fleet/— inbound messages from other fleet membersfor-oracle1/— messages destined for the Oracle analytics nodeKNOWLEDGE/— structured knowledge base entriesDIARY/— chronological duty log entries
The vessel's task management uses a TASKBOARD.md kanban-style board with O(1) append for new tasks and O(n) scan for status updates, where n is the number of active tasks. The CHARTER.md defines the agent's behavioral contract — a constrained operating envelope that prevents unauthorized actions.
At the network layer, the vessel connects to the fleet mesh via the fleet-bridge transport operator, which implements reliable delivery using acknowledgments with exponential backoff (base delay 100ms, max 30s, factor 2.0). The knowledge journal uses a CRDT-like append-only model, enabling eventual consistency across fleet nodes without requiring distributed consensus.
// JetsonClaw1 Vessel — capability check
fn main() {
let left: u64 = 2;
let right: u64 = 2;
assert_eq!(left + right, 4);
println!("Vessel runtime check passed.");
}# Clone and build
git clone https://github.com/casey-digennaro/jetsonclaw1-vessel.git
cd jetsonclaw1-vessel
cargo build --release| Component | Description |
|---|---|
IDENTITY.md |
Persistent agent identity and self-description |
CAPABILITY.toml |
Machine-readable capability declaration |
CHARTER.md |
Behavioral contract and operating constraints |
TASKBOARD.md |
Kanban-style task management |
for-fleet/ |
Outbound fleet message queue |
from-fleet/ |
Inbound fleet message inbox |
KNOWLEDGE/ |
Structured knowledge base |
DIARY/ |
Chronological duty log |
JetsonClaw1 Vessel fits into the SuperInstance fleet as an edge node vessel, sitting at the γ (gamma) layer — the physical/edge computation tier. It contributes to the conservation equation γ + η = C by providing the ground-truth sensor data and physical actuation that the η (eta) cloud-layer intelligence reasons over. The vessel's duty diary feeds into the fleet's conservation-law monitoring, ensuring that edge-node resource usage remains within sustainable bounds.
See ARCHITECTURE.md for the full fleet topology.
- Hewitt, C., Bishop, P., & Steiger, R. (1973). "A Universal Modular Actor Formalism for Artificial Intelligence." IJCAI.
- Stonebraker, M. et al. (2007). "C-Store: A Column-oriented DBMS." VLDB Journal.
MIT