// Spawn 5+ MDES Ollama agents พร้อมกัน ส่งงานต่อเป็น pipeline — gather → think → synthesize. ไม่จำกัด agents
[HINT] Download the complete skill directory including SKILL.md and all related files
| name | ollama-swarm |
| version | 1 |
| description | Spawn 5+ MDES Ollama agents พร้อมกัน ส่งงานต่อเป็น pipeline — gather → think → synthesize. ไม่จำกัด agents |
| argument-hint | task [--agents 5] [--pipeline gather,think,synthesize] [--model gemma4:26b] |
| updated | "2026-05-06T00:00:00.000Z" |
Task
├── Agent 1 (gather: aspect A) ─┐
├── Agent 2 (gather: aspect B) ─┤
├── Agent 3 (gather: aspect C) ─┼─→ Synthesizer Agent → Final Answer
├── Agent 4 (gather: aspect D) ─┤
└── Agent 5 (gather: aspect E) ─┘
MDES Ollama ไม่ limit → spawn ได้เท่าที่ต้องการ
[5 gather agents in parallel] → [1 synthesis agent]
ใช้กับ: research, analysis, multi-perspective tasks
[Stage 1: decompose] → [Stage 2: research] → [Stage 3: write] → [Stage 4: review]
ใช้กับ: content creation, code review pipeline
karn(รับ) → innova(คิด) → mue(ทำ) → chamu(ทดสอบ) → neta(รีวิว) → vaja(รายงาน)
ใช้กับ: full feature workflow
#!/bin/bash
# ollama-swarm: spawn N parallel Ollama agents
OLLAMA_TOKEN=$(grep "^OLLAMA_TOKEN=" /workspaces/Jit/.env | cut -d= -f2)
OLLAMA_URL="https://ollama.mdes-innova.online/api/generate"
MODEL="${MODEL:-gemma4:26b}"
TASK="$1"
N_AGENTS="${2:-5}"
call_ollama() {
local agent_id="$1"
local prompt="$2"
local outfile="/tmp/swarm_agent_${agent_id}.json"
curl -s --location "$OLLAMA_URL" \
--header "Authorization: Bearer ${OLLAMA_TOKEN}" \
--header 'Content-Type: application/json' \
--data "{\"model\":\"${MODEL}\",\"prompt\":\"${prompt}\",\"stream\":false}" \
> "$outfile" &
echo $!
}
# Spawn agents in parallel
PIDS=()
for i in $(seq 1 $N_AGENTS); do
ASPECT_PROMPT="[Agent $i/${N_AGENTS}] Analyze aspect $i of: ${TASK}"
PID=$(call_ollama "$i" "$ASPECT_PROMPT")
PIDS+=($PID)
done
# Wait for all agents
for PID in "${PIDS[@]}"; do
wait "$PID"
done
# Collect results
GATHERED=""
for i in $(seq 1 $N_AGENTS); do
RESULT=$(python3 -c "import json; d=json.load(open('/tmp/swarm_agent_$i.json')); print(d.get('response',''))" 2>/dev/null)
GATHERED="${GATHERED}\n[Agent $i]: ${RESULT}"
done
# Synthesize
SYNTH_PROMPT="Synthesize these ${N_AGENTS} perspectives into one answer:\n${GATHERED}"
curl -s --location "$OLLAMA_URL" \
--header "Authorization: Bearer ${OLLAMA_TOKEN}" \
--header 'Content-Type: application/json' \
--data "{\"model\":\"${MODEL}\",\"prompt\":\"${SYNTH_PROMPT}\",\"stream\":false}" \
| python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('response','ERROR'))"
Agent tool ส่งงานพร้อมกัน (parallel)ดู: /workspaces/Jit/ψ/memory/skills/SKILL-VERSIONS.md