M.I.M.I.R - Multi-agent Intelligent Memory & Insight Repository

# Mimir Open-WebUI Integration This directory contains the **Mimir Planning Assistant Pipeline** for Open-WebUI. ## 🎯 Current Status: Phase 1 Only **Focus**: Proving MCP connection + PM Agent task breakdown This simplified pipeline focuses on: 1. ✅ Testing MCP server connectivity 2. ✅ PM Agent (Ecko) creates task breakdown 3. ✅ Displays TODO list in chat 4. ✅ Saves plan to Neo4j **Not included yet** (Phase 2): - ❌ Multi-agent execution - ❌ Real-time monitoring - ❌ Final report synthesis - ❌ Sidebar updates ## 📂 Files - `mimir_orchestrator.py` - Simplified Phase 1 pipeline - `test_phase1.py` - Connection test script - `README.md` - This file ## 🚀 Quick Start ### 1. Start Services ```powershell # From repo root docker-compose up -d # Wait 30 seconds for services to initialize Start-Sleep -Seconds 30 ``` ### 2. Test MCP Connection (Recommended) ```powershell # Run test script cd pipelines python test_phase1.py ``` Expected output: ``` 🔍 Testing MCP Server health... ✅ MCP Server is running 📚 Listing available MCP tools... ✅ Found 15 tools: - mimir_chain - mimir_execute - todo - todo_list ... 🎯 Testing mimir-chain (PM Agent)... ✅ PM Agent responded successfully ``` ### 3. Access Open-WebUI Open browser: `http://localhost:3000` 1. Create an account (first user is admin) 2. Look for "Mimir: PM Planning Assistant (Phase 1)" in models list 3. Select it and start chatting ### 4. Test PM Agent Try these prompts: ``` Build a REST API with user authentication Create a simple TODO app with React frontend Design a microservices architecture for e-commerce ``` Expected flow: ``` User: Build a REST API with user authentication 🔍 Testing MCP Server Connection... ✅ MCP Connection: OK 🎯 PM Agent (Ecko): Analyzing request and creating task breakdown... ✅ PM Task Breakdown Complete! 📋 TODO List ID: `todoList-abc123` 📊 Tasks Created: 4 tasks ## 📝 Task Plan Summary Breaking project into 4 phases: 1. Research authentication patterns 2. Design database schema 3. Implement API endpoints 4. Write integration tests ## 📋 Task List 1. ⏳ **Research authentication patterns** - Evaluate JWT vs session-based auth 2. ⏳ **Design database schema** - Create users, roles, and sessions tables 3. ⏳ **Implement API endpoints** - POST /auth/login, /auth/logout, /auth/refresh 4. ⏳ **Write integration tests** - Test auth flow end-to-end --- ## 🎯 Next Steps The task plan has been saved to Neo4j. **Options:** 1. Review and edit tasks in Neo4j before execution 2. Run `mimir-execute` to start multi-agent execution (coming in Phase 2) 3. Query the plan: `Show me the tasks in list todoList-abc123` 💡 **Workflow saved at**: 2025-11-05T12:34:56.789Z ``` ## 🎨 UI Features ### Sidebar (Left Panel) ``` ┌──────────────────────────┐ │ 🎯 Mimir Workflow │ │ ⚙️ Executing (45%) │ │ ▓▓▓▓▓░░░░░ │ ├──────────────────────────┤ │ ✅ Research options │ │ └─ ecko │ │ ⚙️ Design schema │ │ └─ worker (active) │ │ ⏳ Implement endpoints │ │ └─ worker │ │ ⏳ Write tests │ │ └─ worker │ ├──────────────────────────┤ │ Active Agents │ │ ● WORKER (2 active) │ │ ● QC (1 active) │ └──────────────────────────┘ ``` ### Main Chat (Right Panel) ``` User: Build a REST API... 🎯 PM Agent: Analyzing request... ✅ PM Summary: Breaking into 4 phases 📋 TODO List ID: todoList-1234 (4 tasks) [PM Agent Full Reasoning] ▸ ⚡ Starting Parallel Execution 🤖 Spawning Worker/QC Agents ✅ Task Completed: Research options (by ecko) 💡 Key Output: JWT with refresh tokens recommended [Task Details: Research options] ▸ ⚙️ Task In Progress: Design schema (by worker) ✅ Task Completed: Design schema (by worker) 💡 Key Output: PostgreSQL with normalized tables --- 📊 Final Report Agent: Synthesizing... # 📊 Mimir Workflow Final Report ... complete summary ... 💾 Saving workflow to memory bank... ✅ Workflow saved: Memory node memory-1-xyz ``` ## ⚙️ Configuration Edit pipeline settings in Open-WebUI: 1. Go to **Workspace → Models** 2. Click "Mimir: PM Planning Assistant (Phase 1)" 3. Adjust **Valves** (settings): | Setting | Default | Description | |---------|---------|-------------| | `MCP_SERVER_URL` | `http://mcp-server:3000` | MCP server endpoint | | `SHOW_PM_FULL_OUTPUT` | `true` | Show complete PM reasoning | | `COLLAPSE_PM_DETAILS` | `true` | Collapse PM output by default | | `TEST_CONNECTION_ON_STARTUP` | `true` | Test MCP on pipeline load | ## 🔧 How It Works ### Connection Test ```python GET {MCP_SERVER_URL}/health → Returns 200 if MCP server is running ``` ### PM Agent Call ```python POST {MCP_SERVER_URL}/message Body: { "method": "tools/call", "params": { "name": "mimir_chain", "arguments": { "task": "User's request", "agent_type": "pm", "create_todos": true } } } → PM analyzes request → Creates TODO list in Neo4j → Returns task breakdown ``` ### Response Parsing ```python # Pipeline extracts: - todo_list_id: "todoList-xyz" - tasks: [{title, description, status}, ...] - summary: "Breaking into N phases..." ``` ## 🐛 Troubleshooting ### Pipeline Not Showing Up in Open-WebUI ```powershell # Check pipeline is mounted docker exec mimir-open-webui ls -la /app/pipelines # Should see: mimir_orchestrator.py ``` **Fix**: Restart Open-WebUI container ```powershell docker-compose restart open-webui ``` ### MCP Connection Failed ```powershell # Check MCP server is running docker ps | Select-String mcp-server # Test from host curl http://localhost:3000/health # Test from inside Open-WebUI docker exec mimir-open-webui curl http://mcp-server:3000/health ``` **Common issues**: 1. MCP server not started: `docker-compose up -d mcp-server` 2. Wrong network: Check `docker-compose.yml` network config 3. Port conflict: Check `docker ps` for port 3000 ### PM Agent Returns ERROR Check MCP server logs: ```powershell docker logs mcp-server --tail 50 ``` Common errors: - `mimir_chain tool not found` → MCP server needs restart - `Neo4j connection failed` → Check Neo4j is running - `Timeout` → Increase timeout in pipeline (default: 300s) ### Empty Task List PM response might not include TODO list ID. Enable full output: 1. Open-WebUI → Workspace → Models 2. Find "Mimir: PM Planning Assistant" 3. Set `SHOW_PM_FULL_OUTPUT` = `true` 4. Set `COLLAPSE_PM_DETAILS` = `false` 5. Try again and inspect full PM output ### Test Manually Run test script: ```powershell cd pipelines python test_phase1.py ``` This tests: - MCP health endpoint - Tool listing - mimir-chain call - Response parsing ## 📚 Architecture ``` User Input in Open-WebUI ↓ mimir_orchestrator.py (Phase 1 Pipeline) ↓ [Test Connection] ↓ MCP Server (HTTP Transport) ↓ mimir-chain tool ↓ PM Agent (Ecko) ↓ Creates TODO list in Neo4j ↓ Returns task breakdown ↓ Pipeline displays in chat ``` **Phase 2 (Future)**: Add Worker/QC agents, real-time monitoring, final reports ## 🎯 Design Decisions 1. **Why simplify to Phase 1?** - Prove MCP connection works first - Test PM agent in isolation - Easier debugging - Build confidence before adding complexity 2. **Why Open-WebUI?** - 114k GitHub stars (production-ready) - Zero UI code needed - Native chat interface - Easy customization via Pipelines 3. **Why Python Pipeline vs Function?** - Pipelines = Workflow orchestration (our use case) - Functions = UI extensions (not needed yet) 4. **Why HTTP transport for MCP?** - Docker-friendly (no stdio complexity) - Easy to test with curl - Works across containers - Standard JSON-RPC protocol ## 🚀 Next Steps (Phase 2) Once Phase 1 is stable: 1. ✅ Add `mimir-execute` call 2. ✅ Implement task monitoring loop 3. ✅ Add real-time progress updates 4. ✅ Create final report synthesis 5. ✅ Save to memory nodes 6. ✅ Add sidebar UI components ## 📖 Related Documentation - [Open-WebUI Pipelines Docs](https://github.com/open-webui/pipelines) - [Mimir Architecture](../docs/architecture/MULTI_AGENT_GRAPH_RAG.md) - [MCP Protocol Spec](https://spec.modelcontextprotocol.io/) - [Integration Summary](../docs/integration/OPENWEBUI_INTEGRATION.md) --- **Version**: 1.0.0-phase1 **Last Updated**: 2025-11-05 **Status**: Phase 1 - PM Agent Testing **Maintainer**: Mimir Development Team