Distributed Computing

AetherShell supports distributed agent execution across multiple nodes with cluster management, job scheduling, and result aggregation.

Cluster Management

Creating a Cluster

cluster_create "my-cluster" { max_nodes: 10, timeout: 30000 }

Adding Nodes

cluster_add_node "my-cluster" { address: "192.168.1.10", port: 3000 }
cluster_add_node "my-cluster" { address: "192.168.1.11", port: 3000 }

Cluster Status

cluster_status "my-cluster"
# {
#   name: "my-cluster",
#   nodes: 2,
#   healthy: 2,
#   total_jobs: 0,
#   uptime_seconds: 120
# }

Node Management

cluster_nodes "my-cluster"
# [
#   { address: "192.168.1.10", port: 3000, status: "active", load: 0.2 },
#   { address: "192.168.1.11", port: 3000, status: "active", load: 0.1 }
# ]

cluster_remove_node "my-cluster" "192.168.1.11"

Job Scheduling

Submitting Jobs

let job_id = job_submit "my-cluster" {
  code: 'ls "src" | map(fn(f) => f.name)',
  priority: "high"
}
echo job_id   # "job-abc123"

Job Status

job_status "job-abc123"
# { id: "job-abc123", status: "running", node: "192.168.1.10", progress: 0.5 }

Job Results

let results = job_results "job-abc123"
echo results

Listing and Canceling

job_list "my-cluster"
# [{ id: "job-abc123", status: "running" }, { id: "job-def456", status: "completed" }]

job_cancel "job-abc123"

Remote Execution

remote_exec is a stub — it does not run anything. There is no SSH/RPC transport behind it. It validates that the node is registered and echoes the request back with status: "simulated" and simulated: true. This page previously showed it returning a real result (# 15), which it never did.

For real remote execution use ssh_exec, which is effect-tagged Exec and approval-gated in agent mode.

remote_exec "192.168.1.10:3000" 'ls "src" | len'
# { node_id: "192.168.1.10:3000", command: "ls \"src\" | len",
#   status: "simulated", simulated: true,
#   output: "remote_exec is a stub: the command was NOT run. …" }

# Actually run it:
ssh_exec "user@192.168.1.10" "ls src | wc -l"

Result Aggregation

Collect and merge results from multiple nodes:

let results = aggregate_results "my-cluster" "job-batch-1"
# Merges results from all nodes into a single value

NANDA Consensus

For coordinated multi-agent decisions, AetherShell provides a consensus protocol:

# Propose a decision
let proposal_id = nanda_propose "Should we deploy v2.0?" {
  options: ["yes", "no", "defer"],
  quorum: 3,
  timeout: 60000
}

# Agents vote
nanda_vote proposal_id "yes" { reason: "All tests pass" }

# Check status
nanda_status proposal_id
# { proposal: "...", votes: 2, quorum: 3, status: "pending" }

# Check if quorum reached
nanda_quorum proposal_id
# false

# Final consensus
nanda_consensus proposal_id
# { decision: "yes", votes_for: 3, votes_against: 0 }

TUI Distributed Panel

The TUI provides a dedicated Distributed Agents tab (Tab 5) for visual management:

• s — Start distributed swarm
• d — Stop distributed swarm
• r — Refresh network status
• t — Test node connections

See TUI Navigation for all key bindings.