Sean's Blog

Personal Assistant

July 4, 2025
Edit on GitHub

A PA powered by LLM via chat and voice interface

  • informs you every day what is happening (daily agenda)
  • executes tasks for you (setup a blog for me, order my favorite pizza, etc.)

Integrations

Instead of having separate apps installed or navigating to specialized websites to order products / food, find events, plan my next travel trip, etc. - we will use one unified chat interface (LLM agent) with integrations for services / data sources that enable your agent to do all these things for you.

Custom Interfaces

Every app / program can be created / customized with natural language

  • create an app for me so I stay motivated to do fitness -> uses personal (emotional) triggers to keep you going
  • add function X to my TODO app

Files

Every file has a semantic embeddings making it discoverable with natural language queries (should be on OS level)

Every media file can be manipulated with natural language

  • decompose a song into its segments i.e. replace or remove the vocals, change the bass line, remove this annoying segment etc.
  • change the style of an image / video
  • create completely new custom videos (including spin-offs of your favorite TV shows)

Ethics

This unified interface has immense abuse potential and thus it is of importance to build a local privacy first version of this - fuck off privacy invading kraken services.

Security

Projects like OpenClaw have shown the first glimpse of LLM powered personal assistant but the security concerns are real: all personal information your personal LLM agent has access to can be potentially exposed via prompt injection and other exploit techniques.

A good agent security layer would let me scope data access for every agent running + ask me via notifications to approve / deny data access in ongoing tasks.

  • https://github.com/botiverse/agent-vault
  • https://github.com/clawshell/clawshell

References

  • https://openclaw.ai/
  • https://github.com/Dicklesworthstone/pi_agent_rust
  • https://github.com/nearai/ironclaw
  • https://www.docker.com/blog/run-nanoclaw-in-docker-shell-sandboxes/
  • https://the-decoder.de/openclaw-aka-clawdbot-und-moltbook-sind-ein-paradies-fuer-datendiebe/
  • https://seksbot.com/

#AI