microdao-daarion/QUICKSTART_PHASE3.md

⚡ QUICKSTART: Phase 3 — LLM + Memory + Tools

One-task start for real agent intelligence

---

🎯 What Phase 3 Adds

Before (Phase 2)	After (Phase 3)
Mock LLM responses	Real GPT-4/DeepSeek/Local
No memory	RAG with vector search
No tools	Tool execution (projects, tasks, etc.)

---

🚀 One-Command Start

# Copy Phase 3 master task
cat docs/tasks/PHASE3_MASTER_TASK.md | pbcopy

# Paste into Cursor AI
# Press Enter
# Wait ~2-3 hours

Cursor will create:

• ✅ llm-proxy (10 files)
• ✅ memory-orchestrator (9 files)
• ✅ toolcore (8 files)
• ✅ docker-compose updates
• ✅ agent-runtime integration

---

🔑 Prerequisites

1. OpenAI API Key (or Local LLM)

Option A: OpenAI

export OPENAI_API_KEY="sk-..."

Option B: Local LLM (Ollama)

# Install Ollama
curl https://ollama.ai/install.sh | sh

# Pull model
ollama pull qwen2.5:8b

# Run server
ollama serve

2. Vector Database

Option A: pgvector (PostgreSQL extension)

CREATE EXTENSION IF NOT EXISTS vector;

Option B: Simple stub (Phase 3 OK)

# Memory Orchestrator can work with simple PostgreSQL
# Vector search = stub for Phase 3

---

📦 After Implementation

Start Services:

# If using existing start script
./scripts/start-phase2.sh  # Existing services

# Start Phase 3 services
docker-compose -f docker-compose.phase3.yml up -d

Test LLM Proxy:

curl -X POST http://localhost:7007/internal/llm/proxy \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4.1-mini",
    "messages": [
      {"role": "user", "content": "Hello!"}
    ],
    "metadata": {
      "agent_id": "agent:sofia",
      "microdao_id": "microdao:daarion"
    }
  }'

# Expected: Real GPT-4 response!

Test Memory:

# Query
curl -X POST http://localhost:7008/internal/agent-memory/query \
  -H "Content-Type: application/json" \
  -d '{
    "agent_id": "agent:sofia",
    "microdao_id": "microdao:daarion",
    "query": "What did we discuss about Phase 3?",
    "limit": 5
  }'

Test Tools:

# List tools
curl http://localhost:7009/internal/tools

# Call tool
curl -X POST http://localhost:7009/internal/tools/call \
  -H "Content-Type: application/json" \
  -d '{
    "tool_id": "projects.list",
    "agent_id": "agent:sofia",
    "microdao_id": "microdao:daarion",
    "args": {}
  }'

---

🧪 E2E Test

In Messenger UI:

User: "Sofia, що нового в проєкті DAARION?"

Expected Agent Flow:

1. ✅ Query memory (past discussions)
2. ✅ Call tool: projects.list
3. ✅ Build prompt with context
4. ✅ Call real LLM (GPT-4)
5. ✅ Post rich reply

Sofia: "В проєкті DAARION є кілька оновлень:

• Phase 2 Agent Integration завершено ✅
• Phase 3 LLM Proxy в розробці 🔄
• Додано 3 нові агенти

Хочете детальніше по якомусь пункту?"

---

📊 Service Status

# Check all Phase 3 services
docker ps | grep -E '(llm-proxy|memory-orchestrator|toolcore)'

# Check health
curl http://localhost:7007/health  # LLM Proxy
curl http://localhost:7008/health  # Memory
curl http://localhost:7009/health  # Toolcore

# Check logs
docker logs -f llm-proxy
docker logs -f memory-orchestrator
docker logs -f toolcore

---

🎯 Success Indicators

After Phase 3:

• ✅ Agent uses real LLM (not keyword mock)
• ✅ Agent remembers conversations
• ✅ Agent can execute tools
• ✅ Responses are intelligent & contextual
• ✅ Latency still < 5s

---

🐛 Troubleshooting

LLM Proxy not working?

# Check API key
docker logs llm-proxy | grep "OPENAI_API_KEY"

# Test provider directly
curl https://api.openai.com/v1/models \
  -H "Authorization: Bearer $OPENAI_API_KEY"

Memory not working?

# Check PostgreSQL connection
docker logs memory-orchestrator | grep "PostgreSQL"

# Check embeddings
docker logs memory-orchestrator | grep "embedding"

Tools not working?

# Check registry loaded
curl http://localhost:7009/internal/tools

# Check permissions
docker logs toolcore | grep "allowed_agents"

---

📚 Documentation

• PHASE3_MASTER_TASK.md — Complete spec
• PHASE3_READY.md — Overview
• PHASE3_ROADMAP.md — Detailed plan

---

🔜 Next Steps

After Phase 3 works:

1. Test with multiple agents
2. Add more tools (task.create, followup.create)
3. Tune memory relevance
4. Optimize LLM costs
5. Monitor usage

---

Time to Start: Copy PHASE3_MASTER_TASK.md into Cursor! 🚀

Questions? Check PHASE3_READY.md first.