Skip to main content

Features Overview - DevoxxGenie Documentation

DevoxxGenie offers a comprehensive set of features designed to enhance your development workflow with AI assistance. This page provides an overview of the key features available in the plugin.

Core Features​

Spec-driven Development (SDD)​

Define tasks as structured markdown specs with acceptance criteria, and let the LLM agent implement them autonomously:

  • Backlog.md Integration: Tasks are stored as version-controlled markdown files with YAML frontmatter
  • Task List & Kanban Board: Browse tasks in a tree view or manage them visually on a drag-and-drop Kanban board with archive-by-drag support
  • Implement with Agent: One-click task implementation — the agent reads the spec, makes code changes, and checks off acceptance criteria as it works
  • 17 Built-in Backlog Tools: Create, edit, search, and complete tasks, documents, and milestones programmatically during agent execution

Multi-Provider LLM Support​

Connect to a wide range of LLM providers:

  • Local Providers: Ollama, LMStudio, GPT4All, Llama.cpp, Jan, and custom OpenAI-compatible providers
  • Cloud Providers: OpenAI, Anthropic (Claude), Mistral, Groq, Google (Gemini), Grok (xAI), DeepInfra, DeepSeek, Kimi, GLM (Zhipu AI), OpenRouter, Azure OpenAI, and Amazon Bedrock

Chat Interface​

  • Streaming responses: See each token as it's received from the LLM in real-time
  • Code highlighting: Proper syntax highlighting for code blocks in responses
  • Chat memory: Configurable conversation history to maintain context
  • Context files: Add files and code snippets to the chat context

Code Completion​

  • Inline Completion: AI-powered code suggestions as you type using Fill-in-the-Middle (FIM) models via Ollama or LM Studio
  • Context-aware: Uses code before and after your cursor for intelligent suggestions
  • Ghost text: Suggestions appear as gray text inline with your code
  • Partial acceptance: Accept full suggestions, single words, or just the current line

Advanced Context Features​

  • RAG Support: Retrieval-Augmented Generation for automatically finding and incorporating relevant code from your project
  • Security Scanning: Detect hardcoded secrets (Gitleaks), SAST issues (OpenGrep) and vulnerable dependencies (Trivy) via LLM agent tools, with automatic backlog task creation from findings
  • DEVOXXGENIE.md: Generate and customize a project description file that gets included in the system prompt
  • Abstract Syntax Tree (AST) context: Automatically include parent class and class/field references

Developer Tools​

  • Inline Completion: AI-powered code completion using Fill-in-the-Middle (FIM) models via Ollama or LM Studio, providing context-aware suggestions as you type
  • Commands: Built-in and custom slash commands (/test, /explain, /review, /find, etc.) with $ARGUMENT placeholder support
  • Skills: LLM-activated SKILL.md capabilities loaded from disk (compatible with .claude/skills and .agents/skills); requires Agent Mode
  • MCP Support: Model Context Protocol servers for extended agent-like capabilities, with a built-in Marketplace for discovering servers
  • Agent Mode (v0.9.4+): Enable agent mode for autonomous codebase exploration using read-only tools. Parallel sub-agents allow concurrent investigation of multiple aspects, each configurable with a different LLM from any provider
  • Web Search: Augment LLM knowledge with web search results from Google or Tavily
  • Token Cost Calculator: Calculate token usage and cost before sending prompts

Multimodal Capabilities​

  • Drag & Drop Images: Add images to your prompts when using multimodal LLMs
  • Multiple File Support: Attach different types of files to your prompts

Feature Matrix by LLM Provider​

FeatureLocal LLMsOpenAIAnthropicGoogleOther Cloud
Chat Interface✅✅✅✅✅
Streaming✅✅✅✅Varies
Spec Driven Dev✅✅✅✅✅
Inline Completion✅*❌❌❌❌
RAG Support✅✅✅✅✅
Project Context✅✅✅✅✅
MCP Support✅✅✅✅✅
Commands✅✅✅✅✅
Skills✅†✅✅✅✅
Web Search✅✅✅✅✅
MultimodalVaries**GPT-4V+Claude 3+Gemini Pro+Varies

* Ollama or LM Studio with FIM-capable models (e.g., starcoder2, qwen2.5-coder)

† Skills require a tool-calling-capable LLM and Agent mode enabled.

**Depends on the model you're using locally (e.g., LLaVA supports images)

Experimental Features​

DevoxxGenie also includes experimental features that are being developed and refined:

  • Test Driven Generation (TDG): Write tests and generate implementation code

Feature Details​

For detailed information about specific features, check out the dedicated pages: