Traditional AI agents burn thousands of tokens re-sending conversation history at every step. Vibra-Ingenn replaces the loop with a single-pass workflow compiler: 1. Your AI writes one JSON recipe (~500 tokens). 2. A native Go runtime executes 1–50 steps locally in milliseconds. 3. The recipe is a file on disk — save it, re-run it, commit it to git, share it. What it does: • MCP server for Claude Desktop, Cursor, Cline, Windsurf, VS Code • REST API on localhost:8990 for any HTTP client • Calls out to any OpenAI-compatible LLM — Ollama, LM Studio, OpenRouter, OpenAI, Anthropic, Gemini, Groq • 7 action types: file, HTTP, script, ai_call, notify, transform, GUI automation • Native Windows mouse/keyboard automation via user32.dll • Conditional DAG branching, retries, cron scheduling, webhook triggers • Encrypted secrets vault, SHA-256 audit receipts, HWID licensing • Ships a custom Ollama Modelfile — `ollama create vibra-llama -f ...` for a purpose-tuned local model Why 3B/7B local models work with it: single-pass compilation eliminates the context bloat that breaks small models in ReAct loops. Vibra also ships a self-healing JSON parser that recovers truncated model output. 7-day free trial. After its $8.99 a month adeptuscamini.com

Vibra-Ingenn (Vibe Engine)

Local AI execution engine — the recipe replaces the agent

Local AI execution engine — the recipe replaces the agent