Skip to content

Latest commit

 

History

History
141 lines (102 loc) · 3.74 KB

CLAUDE_CODE_INSTRUCTIONS.md

File metadata and controls

141 lines (102 loc) · 3.74 KB

🤖 Claude Code Execution Instructions

Razer AIKit Integration Deployment

Priority: HIGH Estimated Time: 30-45 minutes Risk Level: LOW (drop-in replacement, old setup stays as backup)

📋 Task Summary

Deploy Razer AIKit as a Docker sidecar to Visionary Tool Server, adding 22 new MCP tools for local LLM inference with 280K+ HuggingFace models.

This is a DROP-IN REPLACEMENT:

  • Same port 8082
  • Same ngrok domain
  • All existing client configs work unchanged

📁 Files to Deploy

Location: /home/claude/razer-aikit-deployment/ (or provided by user)

  1. docker-compose.aikit.yml → D:\DEV_PROJECTS\GitHub\Claude_Opus_ChatGPT_App_Project\
  2. Dockerfile.aikit → D:\DEV_PROJECTS\GitHub\Claude_Opus_ChatGPT_App_Project\
  3. app/tools/razer_aikit.py → D:\DEV_PROJECTS\GitHub\Claude_Opus_ChatGPT_App_Project\app\tools\
  4. RAZER_AIKIT_DEPLOYMENT.md → D:\DEV_PROJECTS\GitHub\Claude_Opus_ChatGPT_App_Project\

🔧 Execution Steps

Step 1: Create Storage Directories

New-Item -ItemType Directory -Force -Path "D:\Visionary_Models\aikit"
New-Item -ItemType Directory -Force -Path "D:\Visionary_Models\aikit-cache"

Step 2: Copy Deployment Files

Copy all 4 files to their respective locations (listed above).

Step 3: Register AIKit Module

Edit D:\DEV_PROJECTS\GitHub\Claude_Opus_ChatGPT_App_Project\app\server.py:

Add import:

from app.tools import razer_aikit

Step 4: Add HuggingFace Token (if not present)

Check .env.master for HUGGINGFACE_API_KEY. If missing:

HUGGINGFACE_API_KEY=hf_xxxxxxxxxxxx

(User must provide their token from https://huggingface.co/settings/tokens)

Step 5: Build Docker Images

cd D:\DEV_PROJECTS\GitHub\Claude_Opus_ChatGPT_App_Project
docker compose -f docker-compose.aikit.yml build --no-cache

Step 6: Start Stack

docker compose -f docker-compose.aikit.yml up -d

Step 7: Verify Health

# Wait 60 seconds for AIKit to initialize
Start-Sleep -Seconds 60

# Check health endpoints
curl http://localhost:8000/health
curl http://localhost:8082/health

Step 8: Test AIKit Chat

curl -X POST http://localhost:8000/v1/chat/completions `
  -H "Content-Type: application/json" `
  -d '{"model": "deepseek-ai/deepseek-coder-1.3b-instruct", "messages": [{"role": "user", "content": "Hello"}], "max_tokens": 50}'

✅ Success Criteria

  1. docker compose ps shows 2-3 healthy containers
  2. AIKit health returns {"status":"ok"}
  3. Tool Server health shows tools: 334 (312 + 22)
  4. Test chat completion returns valid response
  5. GPU visible inside container (docker exec razer-aikit nvidia-smi)

⚠️ Common Issues

Issue: GPU not detected

Fix: Restart Docker Desktop, ensure WSL2 integration enabled

Issue: Port conflict

Fix: Stop existing services on 8000/8082 or change ports in docker-compose

Issue: Model download slow/fails

Fix: Check HuggingFace token, verify internet connectivity

Issue: Out of memory

Fix: Use smaller model (1.3B instead of 7B)

🔄 Rollback

If anything fails:

docker compose -f docker-compose.aikit.yml down
# Original setup on port 8082 unaffected

📊 Report Back

After execution, provide:

  1. Output of docker compose ps
  2. Output of health checks
  3. Output of test chat completion
  4. Any errors encountered
  5. Tool count from Tool Server health

🎯 Next Steps After Success

  1. All existing clients auto-connect (same port 8082, same ngrok domain)
  2. Test aikit_chat tool via MCP
  3. Try larger models: Qwen/Qwen2.5-7B-Instruct
  4. Enable ngrok tunnel if mobile access needed
  5. Old setup stays on ice as backup (don't delete yet)