JunoStack — Adaptive AI Language Assistant

JunoStack is a custom programming language designed to act as an intelligent, modular assistant for developers. Inspired by RooCode but uniquely built from the ground up, JunoStack supports API key declarations, multi-model routing, assistant roles (modes), and contextual AI workflows using locally hosted models (e.g., Ollama).

---

Features Overview

• ✅ API Endpoint Setup — Define where models are served (e.g., ollama)
• ✅ Mode Assignment — Assign tasks/roles to specific models (with support for friendly aliases)
• ✅ Mode Instructions — Describe behavior per mode (e.g., "debug this")
• ✅ Profile Groups — Group modes into named workflows like production, blueprint, or maintenance (WIP)
• ✅ Dynamic Context Switching — Use currentMode or callMode(...) to invoke AI partners
• ✅ Streaming Ollama Responses — callMode(...) supports live response chunking from Ollama

---

Project Files

• tokenizer.py: token recognition for all language constructs
• parser.py: parses all declarations (apikey, mode, profile, etc.)
• evaluator.py: evaluates instructions, routes prompts, and invokes Ollama with stream support
• runner.py: REPL + script executor
• test_modes2.t: sample demo script for JunoStack's full features

---

Example JunoStack Script

apikey ollama = "http://localhost:11434";

mode coder = "qwen_4b";
modeInstructions["coder"] = "generate code based on user prompts";

mode debug = "gemma_4b";
modeInstructions["debug"] = "explain or refactor problematic code";

profile production {
    coder,
    debug
}

currentProfile = "production"; (WIP)
currentMode = "coder";
callMode("Write a Python function to validate email input.");

---

Aliases Supported

Friendly model names like qwen_4b, gemma_4b, and mistral_small are mapped internally to:

• "qwen_4b" → "qwen3:4b"
• "gemma_4b" → "gemma3:4b"
• "mistral" → "mistral"`

---

How to Run

python runner.py test_modes.t

The interpreter will evaluate your .t file and use callMode(...) to send prompts through the appropriate model.

---

Made by Brandon & Jonah — JunoStack v1.0