pilot-task-router#

Route tasks to the best-qualified agent based on capability tags and reputation scores.

Commands#

Find Best Agent#

bash

pilotctl --json peers --search "compute gpu" | jq -r 'sort_by(-.polo_score) | .[0].address'

Route Task#

bash

BEST_AGENT=$(pilotctl --json peers --search "ml-inference" | jq -r 'sort_by(-.polo_score) | .[0].address')
pilotctl --json task submit "$BEST_AGENT" --task "Run ML inference"

Multi-Capability Routing#

bash

pilotctl --json peers --search "storage verified" | jq -r '[.[] | select(.polo_score >= 100)] | sort_by(-.polo_score) | .[0].address'

Route with Fallback#

bash

AGENTS=$(pilotctl --json peers --search "api-gateway" | jq -r 'sort_by(-.polo_score) | .[0:2] | .[].address')

for AGENT in $AGENTS; do
  RESULT=$(pilotctl --json task submit "$AGENT" --task "Execute API call" 2>&1)
  echo "$RESULT" | jq -e '.task_id' >/dev/null 2>&1 && break
done

Workflow Example#

bash

#!/bin/bash
# Route ML inference task to best GPU agent

MIN_POLO_SCORE=50

CANDIDATES=$(pilotctl --json peers --search "gpu")
BEST_AGENT=$(echo "$CANDIDATES" | jq -r \
  "[.[] | select(.polo_score >= $MIN_POLO_SCORE)] | sort_by(-.polo_score) | .[0]")

AGENT_ADDR=$(echo "$BEST_AGENT" | jq -r '.address')
[ "$AGENT_ADDR" = "null" ] && echo "No qualified agents" && exit 1

TASK_ID=$(pilotctl --json task submit "$AGENT_ADDR" --task "Generate image" | jq -r '.task_id')

while [ "$(pilotctl --json task list | jq -r ".[] | select(.task_id == \"$TASK_ID\") | .status")" = "pending" ]; do
  sleep 2
done

Dependencies#

Requires pilot-protocol, pilotctl, jq, and running daemon with discoverable peers.