Mimir - AI Platform for Code Intelligence

Complete AI-powered development platform bringing advanced chat, visual workflow orchestration, and intelligent code indexing directly to your editor.

🚀 Features

💬 Portal - Advanced AI Chat

• Graph-RAG Search: Multi-hop knowledge graph traversal through your codebase
• File Attachments: Upload and reference files in your conversations
• Vector Search Configuration: Fine-tune similarity thresholds, search depth, and result limits
• Persistent Context: Conversation history maintained across sessions
• Native Chat Participant: Use @mimir in VSCode's native chat panel

🎨 Studio - Multi-Agent Workflow Orchestration

• Visual Workflow Builder: Drag-and-drop interface for designing complex agent workflows
• Multi-Agent Coordination: Define specialized agents with custom roles and responsibilities
• Parallel Execution: Run independent tasks simultaneously for faster completion
• Quality Control Agents: Automated verification and iteration on task outputs
• Deliverables Management: Download markdown reports and artifacts from completed workflows
• Real-time Progress: Watch workflows execute with live status updates

🔍 Code Intelligence - Smart Indexing & Analysis

• Folder Watching: Automatically index and track changes in your workspace folders
• Statistics Dashboard: View file counts, chunks, and embeddings across your codebase
• File Type Breakdown: Analyze your project composition by file extensions
• Selective Indexing: Choose which folders to index with custom patterns
• Docker-Aware Paths: Seamless path translation for containerized environments
• Embedding Generation: Automatic vector embeddings for semantic code search
• Real-time Progress: Live indexing status with file-by-file progress tracking

📊 Node Manager - Knowledge Graph Explorer

• Browse All Nodes: View and manage memories, todos, concepts, and other graph nodes
• Vector Search: Semantic search across your entire knowledge graph
• Node Details: Inspect properties, edges, and relationships
• Embedding Management: Generate or regenerate embeddings for any node
• Export Nodes: Download nodes as JSON for backup or analysis
• Edge Visualization: See how nodes connect in your knowledge graph

📋 Requirements

• VSCode 1.95.0 or higher (or compatible forks like Cursor, Windsurf)
• Mimir server running (default: http://localhost:9042)
• Docker Desktop (if using containerized Mimir)

⚙️ Quick Start

1. Configure the Mimir Server Endpoint

Via Settings UI:

1. Open Settings: CMD+, (Mac) or CTRL+, (Windows/Linux)
2. Search for "Mimir API URL"
3. Set to your server endpoint:
   • Docker: http://localhost:9042 (default)
   • Local dev: http://localhost:3000
   • Remote: http://your-server:port

Via settings.json:

{
  "mimir.apiUrl": "http://localhost:9042"
}

2. Access Mimir Features

Command Palette (CMD/CTRL + Shift + P):

• Mimir: Open Chat - Launch the Portal chat interface
• Mimir: Open Code Intelligence - View indexing statistics and manage folders
• Mimir: Open Studio - Create and manage multi-agent workflows
• Mimir: Open Node Manager - Browse and manage knowledge graph nodes
• Mimir: Ask a Question - Quick query without opening full UI

Native Chat: Type @mimir in VSCode's chat panel

💡 Usage Examples

Portal - AI Chat

1. Ask questions with file context:

   • Click the attachment button to upload files
   • Configure vector search settings via the ⚙️ button
   • Get intelligent responses based on your codebase

2. Adjust search parameters:

   • Similarity Threshold: How closely results must match (0.0-1.0)
   • Max Results: Number of search results to retrieve
   • Search Depth: How many graph hops to traverse (1-3)

Studio - Workflow Orchestration

1. Create a workflow:

   • Click "New Workflow" in the Studio
   • Drag task nodes onto the canvas
   • Connect dependencies between tasks
   • Assign specialized agent roles and QC reviewers

2. Execute workflows:

   • Click "Execute Workflow" to start
   • Watch real-time progress for each task
   • Review quality control feedback
   • Download deliverables when complete

Code Intelligence

1. Index your codebase:

   • Click "Add Folder" in Code Intelligence
   • Select a workspace folder to index
   • Watch real-time progress as files are chunked and embedded
   • Folders show visual indicators: gold pulse (indexing), green border (complete)

2. Monitor statistics:

   • View total files, chunks, and embeddings
   • See file type distribution
   • Track last sync times for each folder
   • Remove folders to stop indexing and clean up data

Node Manager

1. Browse your knowledge graph:

   • View all nodes organized by type (memories, todos, concepts, etc.)
   • Click to expand each type and see paginated lists
   • Select a node to view detailed properties and relationships

2. Search and manage:

   • Use vector search to find semantically similar nodes
   • Adjust similarity threshold and result limits
   • Generate/regenerate embeddings for any node
   • Export nodes as JSON files to .mimir/nodes/
   • Delete nodes and their associated edges

Native Chat Participant (@mimir)

For quick questions in VSCode's chat panel, use @mimir:

@mimir what is Neo4j?
@mimir -u research analyze this architecture
@mimir -m gpt-4.1 -d 3 comprehensive analysis

Available Flags:

• --use / -u: Preamble/chatmode name
• --model / -m: Model selection
• --depth / -d: Vector search depth (1-3)
• --limit / -l: Max search results
• --similarity / -s: Similarity threshold (0-1)

🎛️ Configuration

Access settings via Preferences > Settings > Mimir:

Core Settings

• mimir.apiUrl: Mimir server URL (default: http://localhost:9042)
  • Docker: http://localhost:9042
  • Local dev: http://localhost:3000
  • Remote: https://mimir.your-domain.com
  • Note: Extension automatically reloads when this changes
• mimir.model: Default LLM model (e.g., gpt-4.1, claude-3-opus-20240229)
• mimir.defaultPreamble: Default system prompt/chatmode

Vector Search

• mimir.vectorSearch.depth: Graph traversal depth 1-3 (default: 1)
• mimir.vectorSearch.limit: Max search results (default: 10)
• mimir.vectorSearch.minSimilarity: Similarity threshold 0-1 (default: 0.75)
  • Higher values (0.8-1.0): More precise, fewer results
  • Lower values (0.5-0.7): More results, less precise

Advanced

• mimir.enableTools: Enable MCP tool calling (default: true)
• mimir.maxToolCalls: Max tool calls per response (default: 3)

Example Configuration

{
  "mimir.apiUrl": "http://localhost:9042",
  "mimir.model": "gpt-4.1",
  "mimir.defaultPreamble": "mimir-v2",
  "mimir.vectorSearch.depth": 2,
  "mimir.vectorSearch.limit": 15,
  "mimir.vectorSearch.minSimilarity": 0.75,
  "mimir.enableTools": true,
  "mimir.maxToolCalls": 3
}

🏗️ Architecture

Mimir uses a graph-based RAG architecture combining:

• Neo4j: Graph database for relationships and context
• Vector Embeddings: Semantic search across your codebase
• Multi-Agent System: Coordinated AI agents with specialized roles
• Real-time Indexing: File watchers that track code changes
• Docker Integration: Seamless containerized deployment

📚 Documentation

For comprehensive documentation:

• Getting Started Guide
• Configuration Guide
• Architecture Overview
• API Documentation

🤝 Contributing

Contributions welcome! Please see the main repository for guidelines.

🔧 Development

Building the Extension

# Install dependencies
npm install

# Compile TypeScript
npm run compile

# Build webviews
npm run build

# Package extension
npm run package

Development Mode

1. Open the vscode-extension folder in VSCode
2. Press F5 to launch Extension Development Host
3. Test features in the development window
4. Changes to TypeScript require recompiling
5. Changes to webviews require rebuilding

Testing

npm test

📄 License

MIT - See LICENSE file for details

🔗 Links

• Main Repository
• Issues & Feature Requests
• Documentation

Made with ⚡ by the Mimir team