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

# Open-WebUI Pipeline Configuration Guide **Version:** 1.0.0 **Last Updated:** 2025-11-10 This guide explains how to configure and use the pre-packaged Mimir pipelines in Open-WebUI, including how to enable them and switch between different LLM providers. --- ## Table of Contents 1. [Overview](#overview) 2. [Pre-Packaged Pipelines](#pre-packaged-pipelines) 3. [First-Time Setup](#first-time-setup) 4. [Enabling Pipelines](#enabling-pipelines) 5. [Switching LLM Providers](#switching-llm-providers) 6. [Pipeline Configuration](#pipeline-configuration) 7. [Troubleshooting](#troubleshooting) --- ## Overview Mimir comes with **4 pre-packaged pipelines** that are automatically copied into your Open-WebUI container: 1. **Mimir Multi-Agent Orchestrator** - Full PM → Worker → QC orchestration 2. **Mimir RAG Auto** - Semantic search with automatic context enrichment 3. **Mimir File Browser** - Browse and search indexed files 4. **Mimir Tools Wrapper** - Command shortcuts (`/list_folders`, `/search`, etc.) These pipelines are **pre-installed** but must be **manually enabled** on first use. --- ## Pre-Packaged Pipelines ### Pipeline Locations After building the Docker image, pipelines are located at: ``` /app/backend/data/pipelines/ ├── mimir_orchestrator.py # Multi-agent orchestration ├── mimir_rag_auto.py # RAG with semantic search ├── mimir_file_browser.py # File browsing actions └── mimir_tools_wrapper.py # Command wrapper filter ``` ### Pipeline Types | Pipeline | Type | Purpose | |----------|------|---------| | **mimir_orchestrator.py** | Pipe | Full multi-agent workflow (Ecko → PM → Workers → QC) | | **mimir_rag_auto.py** | Pipe | Chat with semantic search context enrichment | | **mimir_file_browser.py** | Action | File browsing and statistics | | **mimir_tools_wrapper.py** | Filter | Command shortcuts for tools | --- ## First-Time Setup ### Step 1: Build the Docker Image The pipelines are automatically copied during the Docker build: ```bash # Build the Open-WebUI image with Mimir pipelines docker compose build open-webui # Or rebuild everything npm run build:docker docker compose up -d --build ``` ### Step 2: Verify Pipelines Are Copied Check that pipeline files are in the container: ```bash # List pipeline files docker exec -it mimir-open-webui ls -la /app/backend/data/pipelines/ # Should show: # mimir_orchestrator.py # mimir_rag_auto.py # mimir_file_browser.py # mimir_tools_wrapper.py ``` ### Step 3: Access Open-WebUI ``` http://localhost:3000 ``` **First-time setup:** 1. Create an admin account (first user becomes admin) 2. You'll be redirected to the chat interface --- ## Enabling Pipelines Pipelines are **pre-installed** but **disabled by default**. You must enable them manually. ### Method 1: Via Admin Panel (Recommended) 1. **Open Admin Panel**: - Click your username (bottom-left) - Select **"Admin Panel"** 2. **Navigate to Pipelines**: - Click **"Admin Panel"** in the sidebar - Select **"Pipelines"** tab 3. **Enable Mimir Pipelines**: You should see 4 Mimir pipelines listed: - ✅ **Mimir Multi-Agent Orchestrator** - ✅ **Mimir RAG Auto** - ✅ **Mimir File Browser** - ✅ **Mimir Tools Wrapper** 4. **Toggle Each Pipeline**: - Click the **toggle switch** next to each pipeline to enable it - The switch should turn green/blue when enabled 5. **Verify**: - Go back to the chat interface - You should now see the pipelines available in model selection ### Method 2: Via API (Advanced) If you have an API key, you can enable pipelines programmatically: ```bash # Get your API key from: Settings → Account → API Keys # Enable a pipeline curl -X POST http://localhost:3000/api/v1/pipelines/enable \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "pipeline_id": "mimir_orchestrator" }' ``` --- ## Switching LLM Providers All Mimir pipelines support switching between **Copilot API** (default) and **Ollama** (local). ### Default Configuration (Copilot API) Out of the box, all pipelines use GitHub Copilot API: ```python COPILOT_BASE_URL: str = Field( default="http://copilot-api:4141/v1", description="Copilot API base URL" ) PM_MODEL: str = Field(default="gpt-4.1") WORKER_MODEL: str = Field(default="gpt-4.1") QC_MODEL: str = Field(default="gpt-4.1") ``` ### Switching to Ollama #### Option 1: Via Pipeline Valves (Easiest) 1. **Open Admin Panel** → **Pipelines** 2. **Click the pipeline** you want to configure (e.g., "Mimir Multi-Agent Orchestrator") 3. **Click the gear icon** (⚙️) to edit Valves 4. **Modify the configuration**: ```json { "COPILOT_BASE_URL": "http://ollama:11434/v1", "PM_MODEL": "llama3.1:8b", "WORKER_MODEL": "llama3.1:8b", "QC_MODEL": "llama3.1:8b" } ``` 5. **Save** and test with a new chat #### Option 2: Edit Pipeline Source (Persistent) 1. **Edit the pipeline file** in your local repository: ```bash # Edit the orchestrator pipeline nano pipelines/mimir_orchestrator.py ``` 2. **Change the Valves defaults**: ```python class Valves(BaseModel): # Change Copilot API URL to Ollama COPILOT_BASE_URL: str = Field( default="http://ollama:11434/v1", # Changed description="Ollama API base URL" ) # Change models to Ollama models PM_MODEL: str = Field( default="llama3.1:8b", # Changed description="Model for PM agent" ) WORKER_MODEL: str = Field( default="llama3.1:8b", # Changed description="Model for worker agents" ) QC_MODEL: str = Field( default="llama3.1:8b", # Changed description="Model for QC agents" ) ``` 3. **Rebuild and restart**: ```bash docker compose build open-webui docker compose up -d --no-deps open-webui ``` ### Hybrid Configuration (Advanced) You can use different providers for different agents: ```python # Use Copilot for PM and QC (strategic) PM_MODEL: str = Field(default="gpt-4o") QC_MODEL: str = Field(default="gpt-4o") # Use Ollama for Workers (high volume) WORKER_MODEL: str = Field(default="llama3.1:8b") ``` > ⚠️ **Limitation**: Currently, all agents must use the same API endpoint. Hybrid setups require code modifications to switch base URLs per agent. --- ## Pipeline Configuration ### Mimir Multi-Agent Orchestrator **Purpose**: Full multi-agent workflow with PM, Workers, and QC agents **Key Configuration Options**: ```python # Agent Enablement ECKO_ENABLED: bool = True # Enable prompt architect PM_ENABLED: bool = True # Enable planning stage WORKERS_ENABLED: bool = True # Enable worker execution # Model Selection PM_MODEL: str = "gpt-4.1" # PM agent model WORKER_MODEL: str = "gpt-4.1" # Worker agent model QC_MODEL: str = "gpt-4.1" # QC agent model # Context Enrichment SEMANTIC_SEARCH_ENABLED: bool = True SEMANTIC_SEARCH_LIMIT: int = 10 ``` **Usage**: 1. Select "Mimir Multi-Agent Orchestrator" as your model 2. Type your request (e.g., "Build a user authentication system") 3. The pipeline will: - Optimize your prompt (Ecko) - Create a task plan (PM) - Execute tasks (Workers) - Verify output (QC) - Generate final report ### Mimir RAG Auto **Purpose**: Chat with automatic semantic search context enrichment **Key Configuration Options**: ```python # LLM Backend MIMIR_DEFAULT_PROVIDER: str = "copilot" # "openai", "copilot", or "ollama" # Model Selection MIMIR_DEFAULT_MODEL: str = "gpt-4.1" # Semantic Search SEMANTIC_SEARCH_ENABLED: bool = True SEMANTIC_SEARCH_LIMIT: int = 10 # Embedding Configuration MIMIR_EMBEDDINGS_MODEL: str = "mxbai-embed-large" ``` **Usage**: 1. Select "Mimir RAG Auto" as your model 2. Ask questions about your indexed codebase 3. The pipeline automatically enriches context with relevant files ### Mimir File Browser **Purpose**: Browse and get statistics on indexed files **Actions Available**: - `list_watched_folders` - List all watched folders - `get_folder_stats` - Get statistics for a specific folder **Usage**: 1. In the chat interface, click the **"+"** button 2. Select **"Actions"** 3. Choose **"Mimir File Browser"** 4. Select an action (e.g., "list_watched_folders") ### Mimir Tools Wrapper **Purpose**: Command shortcuts for common operations **Commands Available**: - `/list_folders` - List watched folders - `/folder_stats <path>` - Get folder statistics - `/search <query>` - Semantic search - `/orchestration <id>` - Get orchestration details **Usage**: 1. Enable "Mimir Tools Wrapper" in Pipelines 2. In any chat, type a command (e.g., `/list_folders`) 3. The command is intercepted and executed directly --- ## Troubleshooting ### Problem: Pipelines Not Showing Up **Symptoms**: No Mimir pipelines visible in Admin Panel **Solution**: ```bash # 1. Verify files are in container docker exec -it mimir-open-webui ls -la /app/backend/data/pipelines/ # 2. Check Open-WebUI logs docker compose logs open-webui | grep -i pipeline # 3. Rebuild the image docker compose build open-webui --no-cache docker compose up -d --force-recreate open-webui ``` ### Problem: Pipeline Fails to Enable **Symptoms**: Toggle switch doesn't stay enabled **Solution**: ```bash # 1. Check for Python syntax errors docker exec -it mimir-open-webui python3 -m py_compile \ /app/backend/data/pipelines/mimir_orchestrator.py # 2. Check dependencies are installed docker exec -it mimir-open-webui pip list | grep -E "neo4j|aiohttp|pydantic" # 3. Check Open-WebUI logs for errors docker compose logs open-webui | tail -50 ``` ### Problem: Pipeline Can't Connect to Neo4j **Symptoms**: Error about Neo4j connection **Solution**: ```bash # 1. Verify Neo4j is running docker compose ps neo4j # 2. Check Neo4j connection from Open-WebUI container docker exec -it mimir-open-webui ping neo4j_db # 3. Verify Neo4j credentials in pipeline Valves: # NEO4J_URL: bolt://neo4j_db:7687 # NEO4J_USER: neo4j # NEO4J_PASSWORD: password ``` ### Problem: Pipeline Can't Connect to Copilot API **Symptoms**: Error about Copilot API connection **Solution**: ```bash # 1. Verify Copilot API is running docker compose ps copilot-api # 2. Check Copilot API from Open-WebUI container docker exec -it mimir-open-webui curl http://copilot-api:4141/v1/models # 3. Verify Copilot API URL in pipeline Valves: # COPILOT_BASE_URL: http://copilot-api:4141/v1 ``` ### Problem: Ollama Models Not Found **Symptoms**: Error: Model 'llama3.1:8b' not found **Solution**: ```bash # 1. Enable Ollama service (uncomment in docker-compose.yml) docker compose up -d ollama # 2. Pull the model docker exec -it ollama_server ollama pull llama3.1:8b # 3. Verify model is available docker exec -it ollama_server ollama list ``` ### Problem: Pipeline Changes Not Taking Effect **Symptoms**: Modified Valves but behavior unchanged **Solution**: ```bash # 1. Refresh the page (Ctrl+Shift+R) # 2. Clear browser cache # 3. Restart Open-WebUI docker compose restart open-webui # 4. If editing source files, rebuild docker compose build open-webui --no-cache docker compose up -d --force-recreate open-webui ``` --- ## Best Practices ### 1. Start with Defaults Use the default Copilot API + gpt-4.1 configuration until you identify specific needs. ### 2. Enable Pipelines Incrementally Don't enable all 4 pipelines at once. Start with: 1. **Mimir RAG Auto** - For general chat with context 2. **Mimir Tools Wrapper** - For command shortcuts 3. **Mimir Multi-Agent Orchestrator** - For complex tasks ### 3. Test Configuration Changes Always test Valve changes with a simple query before running complex orchestrations. ### 4. Monitor Resource Usage - **Copilot API**: Monitor usage at http://localhost:4141/usage - **Ollama**: Monitor GPU/RAM usage with `docker stats ollama_server` ### 5. Keep Pipelines Updated When updating Mimir: ```bash # Pull latest changes git pull # Rebuild with updated pipelines docker compose build open-webui --no-cache docker compose up -d --force-recreate open-webui ``` --- ## Summary | Task | Method | |------|--------| | **Enable pipelines** | Admin Panel → Pipelines → Toggle switches | | **Switch to Ollama** | Edit Valves → Change COPILOT_BASE_URL and models | | **Update pipelines** | Edit source → Rebuild → Restart | | **Troubleshoot** | Check logs → Verify connections → Rebuild if needed | --- ## Related Documentation - **[QUICKSTART.md](../getting-started/QUICKSTART.md)** - Initial setup - **[LLM_PROVIDER_GUIDE.md](LLM_PROVIDER_GUIDE.md)** - Detailed LLM configuration - **[AGENTS.md](../AGENTS.md)** - Multi-agent workflows - **[Open-WebUI Docs](https://docs.openwebui.com/)** - Official documentation --- **Need Help?** - 🐛 [Report Issues](https://github.com/orneryd/Mimir/issues) - 💬 [Discussions](https://github.com/orneryd/Mimir/discussions)