LocalAgent – Autonomous LLM Tool Runtime

LocalAgent is a Python-based autonomous agent runtime that runs on your local machine, integrates with local LLMs via Ollama, and intelligently decides when to use tools to answer user prompts — inspired by agentic frameworks like AutoGPT, OpenDevin, and Devika.

This project is the foundation for a larger open-source platform for LLM prompt evaluation, runtime orchestration, and agentic behavior testing.

---

Current Features

• Multi-step agent reasoning loop — rethinks after every tool use
• Modular Tool Interface — easy to add tools like shell commands, calculator, etc.
• Local LLM inference via Mistral running on Ollama
• Secure evaluation sandbox for math expressions (eval w/ safelist)
• Learns from prior tool results using an internal history memory
• Handles malformed LLM output (extra lines, hallucinated answers, etc.)

---

Current Stack

Area	Technology
LLM Runtime	Ollama + Mistral
Agent Logic	Python
Tools Framework	Custom Tool class
Math Sandbox	Python eval() w/ whitelist
Shell Access	subprocess.run

---

Example Usage

$ python agent_brain.py

Enter Prompt: what is 52 * 12
[LLM OUTPUT]
TOOL: calc
INPUT: 52 * 12

Jarvis: Result: 624

---

Enter Prompt: create folder test123 and list it
[LLM OUTPUT]
TOOL: shell
INPUT: mkdir test123 && ls test123

Jarvis: a.txt
b.txt
c.txt ...