Back to feed
Dev.to
Dev.to
5/10/2026
I Built an AI Agent Orchestrator Where Gemma 4 Only Knows What You Teach It

I Built an AI Agent Orchestrator Where Gemma 4 Only Knows What You Teach It

Short summary

GemmaOrch is a skill-based AI agent orchestrator that uses Gemma 4 to answer only within scope defined by Markdown skill files—refusing out-of-scope queries without hallucination. The architecture separates expertise (plain Markdown files) from orchestration logic (Java/Spring), supports GitHub skill imports, streaming chat, MCP integration for IDEs, and self-hosting via Ollama. Open-source project demonstrating why the 31B dense model beats smaller variants for constraint-adherent reasoning.

  • •Skill files define agent scope; model refuses off-topic queries reliably
  • •Separates expertise (Markdown) from code (Java/Spring), simplifying maintenance
  • •Integrates with Claude Code and Cursor as MCP tools; self-hostable via Ollama

Generated with AI, which can make mistakes.

#ai-agents#ai-tools#open-source
Read full article at Dev.to

Is this a good recommendation for you?

Explore more

I Built a Local-First VSCode Code Mentor with Gemma 4 — Your Code Never Leaves Your Machine
Dev.to
My Local Copilot: Gemma 4 + Open WebUI + OpenHands for Coding Without Leaving My Machine
Dev.to
Local-First AI Done Right: How Gemma 4 E2B and 'Thinking Mode' Powered DiagramFlowAI
Dev.to
Open-Rosalind: Building a Gemma 4-Powered Bio-Agent for Reproducible Life Science Research
Dev.to