I’ve been experimenting with a different direction for personal AI:

not cloud chatbots,

not another wrapper,

but a local-first cognitive operating system.

So I built EIDOLON OS.

An experimental AI system that combines memory, vision, semantic retrieval, CCTV intelligence, workflow replay, and local agent actions into one modular platform.

EIDOLON OS

A local-first AI cognitive operating system designed to transform raw desktop activity into structured, searchable memory.

Current Capabilities

• AI memory engine


• PDF intelligence


• Voice ingestion


• YOLO/OpenCV vision pipelines


• CCTV/video analysis


• Replayable activity timelines


• Agent workflows


• Semantic search


• Daily activity summaries


• Temporal memory graphs

Dashboard

[UPLOAD SCREENSHOT HERE]

Vision / CCTV Intelligence

EIDOLON can analyze uploaded videos and camera feeds using local computer vision pipelines.

Features include:

• motion detection


• object detection


• YOLO vision analysis


• event timelines


• temporal memory storage

[UPLOAD SCREENSHOT HERE]

Agent Workflows

The system also includes a local action/agent layer capable of:

• opening applications


• summarizing activity


• replaying sessions


• workflow intelligence


• contextual memory actions

[UPLOAD SCREENSHOT HERE]

Architecture

Built with:

• Next.js


• FastAPI


• TypeScript


• Python


• OpenCV


• YOLO

Everything runs locally.

No cloud dependency.

No telemetry.

No external memory APIs.

Why I Built This

Most AI systems today are stateless chat interfaces.

I wanted to explore something different:

an AI system that continuously remembers, observes, structures, and retrieves contextual information like a cognitive layer for the machine itself.

Still early.

But the foundation is becoming real.

GitHub