What it does

DevoxxGenie is a Java-based IntelliJ plugin that wires local LLMs (Ollama, LMStudio, llama.cpp, Exo) and cloud providers (OpenAI, Anthropic, Gemini, Groq, Azure, Bedrock, and others) into your IDE. It started as a code explainer/reviewer, but has grown into something closer to a project operating system: task management, autonomous agents, inline completion, security scanning, and even a protocol for talking to external CLI tools like Claude Code and Copilot.

The interesting bit

The Spec Driven Development (SDD) workflow is the unusual part. You write tasks in Backlog.md, browse them in a Kanban board inside IntelliJ, then hit “Implement with Agent” and let the LLM work through them in dependency order via an “Agent Loop.” Security findings from Gitleaks, OpenGrep, and Trivy automatically become backlog tasks too. There’s even a reflection-based ExternalPromptService so other IntelliJ plugins can feed it findings without compile-time coupling — demonstrated with SonarLint and SpotBugs forks.

Key highlights

• Broad LLM support: 15+ providers including local inference tools and major cloud APIs
• SDD with Agent Loop: Kanban task board, topological batch execution, and autonomous implementation
• Security scanning as agent tools: Gitleaks (secrets), OpenGrep (SAST), Trivy (CVEs) — findings auto-convert to backlog tasks
• ACP/CLI Runners: JSON-RPC 2.0 protocol to orchestrate external agents (Claude Code, Gemini CLI, Kimi, Codex, Copilot)
• Portable skills: SKILL.md files shared with Claude Code, Codex, and Gemini ecosystems
• Inline FIM completion: context-aware suggestions via Ollama or LM Studio
• 60K+ downloads, actively maintained with extensive documentation at genie.devoxx.com

Caveats

• The README is enthusiastic about “paradigm shifts” and “the future of agentic programming”; the actual feature set is solid but the hype-to-substance ratio is tilted
• Some newer features (ACP Runners, Plugin Integration API) are documented but the real-world stability is unclear from the sources
• Security scanners require separate local installation (brew install) and manual path configuration

Verdict

Worth a look if you want LLM assistance tightly integrated into your IDE workflow rather than bouncing between browser tabs and terminal windows. Probably overkill if you just want Copilot-style autocomplete. Java developers and teams already using IntelliJ will get the most mileage; VS Code users need not apply.