I Haunted My Own Terminal Using Local Open-Weight LLMs and Ollama

Description and use

Daemon turns a standard terminal into an interactive, spooky environment by injecting AI‑generated commentary and visual anomalies, all while preserving privacy through local inference.

Problem

Create a reactive, privacy‑preserving desktop overlay that injects AI‑generated commentary and visual anomalies into terminal sessions without relying on cloud services.

Outcome

Daemon transforms a standard terminal into an interactive, spooky environment by injecting AI‑generated commentary and visual anomalies, all while preserving privacy through local inference.

Workflow

• Install Daemon and its dependencies (Python, PySide6, psutil, Ollama/llama.cpp).
• Configure optional Ollama integration for context‑aware whispers.
• Run Daemon as a background service; it hooks into the shell via subprocess streams.
• Daemon monitors system telemetry (process names, load, idle state) using psutil.
• When telemetry triggers a condition, the orchestrator selects an effect (overlay, sound, auto‑typing).
• If Ollama is available, generate context‑aware text; otherwise use procedural fallback.
• Inject generated output into the terminal stream or display visual overlays via a transparent Qt canvas.
• Daemon continues to monitor and react in real time, maintaining low latency.
• Stop Daemon when no longer needed.

Evidence

Daemon is a local Python application that turns desktop telemetry into atmospheric audio and visual anomalies.

Source

Open-Weight Models: Leverages Qwen2.5-Coder for context-aware shell command manipulation and syntax-valid auto-typing generation, alongside lightweight open-weight vision/audio classifiers for ambient perception.

Source

Zero cloud APIs or external data services are used—the core intelligence relies strictly on local inference to ensure complete privacy and real-time execution.

Source

With Daemon running silently in the background, I was debugging a project and typed git status into my shell. Instead of the usual dry output, the screen paused for a fraction of a second before streaming out: ...

Source