Menu

1 / 2

I built a self-hosted PC automation system with local LLaMA — it verifies actions actually worked

DEV Community·Hanzzel Corp·7 months ago
#MjsnPV8V
#python#node#ai#blue#arrow
Long article (357 words) — no AI summary yet.
Reading 0:00
15s threshold

The problem I wanted to solve

Every automation tool I tried had the same issue: fire and forget. It sends the command and hopes for the best. I wanted something that verifies the action actually worked.

So I built Blue Arrow — a modular PC automation system that runs entirely local, no cloud, no subscriptions.

How it works

You send a command via Telegram:

"open Chrome" / "search report.pdf" / "write a letter in LibreOffice"

The system processes it through a state machine:

IDLE → INTENT → PLANNING → EXECUTING → VERIFYING → COMPLETED

After every action, a Python Verifier Engine checks if it actually worked:

Detects the process PID

Detects the window ID via wmctrl/xdotool

Checks focus state

Returns a confidence score (0.0–1.0)

If confidence < 0.5, it retries or reports failure.

The AI layer

Local LLaMA via Ollama handles everything:

Intent parsing — interprets what you actually meant

Text generation — drafts letters, explains errors, generates code

Semantic memory — vector embeddings,…

Continue reading — create a free account

Join HashtagPLUS to read full articles, follow hashtags, vote, and join the conversation.