Aigentlm by AeteX
GitHub
ON-DEVICE AGENT • DESKTOP UI RUNTIME
// ZERO EGRESS

An AI agent that lives on your machine, not in someone else's cloud.

Aigentlm pairs a local quantized model with a high-framerate desktop interface. It executes code, manages your file systems, and automates multi-step workflows with Cursor-like velocity without sending a single token or byte off your device.

AeteX Ecosystem ↗
YOUR LOCAL MACHINE • HARDWARE SANDBOX
editing app.py
local model • 0ms egress
organizing downloads
02 // CAPABILITIES

Coding was only the starting point.

Terminal-only local agents are bottlenecked by CLI boundaries. Aigentlm pairs an on-device neural engine with a responsive canvas to perceive and manipulate your desktop workspace.

01 // CODE Llama / Qwen

Write & Refactor

Indexes your codebase locally, applies multi-file diffs, monitors test outputs, and repairs stack traces in a live sandbox.

AST Parsing Local
02 // FILES OS Native

Organize & Batch Clean

Categorizes downloads, normalizes filename conventions, restructures deep folder trees, and cleans duplicate assets.

Disk I/O Sandboxed
03 // BROWSER Headless CDP

Local Research

Spins up an isolated Chromium session to inspect documentation, scrape tables, and distill data directly into your local notes.

Client State Isolated
04 // SYSTEM Daemon Loop

Automate Tasks

Schedules local cron automations, transforms telemetry logs, and triggers actions without third-party webhooks or servers.

Kernel Hooks 0 Egress
03 // ARCHITECTURE & ROADMAP

Engineered for physical hardware sovereignty.

Building in public under the AeteX ecosystem alongside nova voice AI. Here is what is confirmed for the core engine stack.

LLM

Local Weight Ingest

GGUF / EXL2 / Ollama Runtime

Direct GPU metal/CUDA acceleration. Runs Q4/Q8 quantized models locally on MacBook M-series silicon or NVIDIA RTX cards.

IPC

Deterministic Desktop GUI

Rust Core • High FPS

Ultra-lightweight rendering loop with low memory overhead. Real-time visual timeline, task execution canvas, and diff inspector.

BOX

Air-Gapped Sandbox

Zero Telemetry • Strict Local

System calls require explicit permission gates. Hardwired firewall keeps all embeddings, code index, and chat history strictly offline.

Current State: Engine Architecture & Quantized Core
Project Origin: Ai + Agent + LM • Under development by Tawhid Rahman / AeteX
TARGET RELEASE: LATE 2028
04 // WHERE IT FITS

A distinct architectural quadrant.

Positioned against local-agent CLIs and proprietary cloud ecosystems. Aigentlm combines the accessibility of a GUI with pure local sovereignty.

Tool Architecture Runs Locally Full Desktop UI Beyond Coding Tasks Privacy Model
Aigentlm Yes • 100% on-device Yes • Native Desktop Yes • Files, OS, Web Zero Egress Air-Gap
Cursor No • Cloud Hosted Yes • VS Code Fork No • Coding Only Proprietary Cloud API
Cline / Terminal Agents Yes • via Ollama/vLLM No • CLI / Terminal Only No • Coding Centric Local or Custom API
Cloud Agent Platforms No • Remote VM Yes • Web App Yes • General Web Complete Cloud Telemetry
05 // EARLY ACCESS

Join the local runtime waitlist.

Aigentlm is in active development with a release window set for early 2027. Early engineering builds will be shared with contributors and tester cohorts.

// Direct Channel PGP Ready

For collaboration, hardware testing partnerships, or questions regarding Aigentlm or nova voice AI:

aetexhq@proton.me
© 2026–2027 Aigentlm • Part of AeteX • By Tawhid Rahman