Personal AI Agents

Exploring autonomous agents for research and productivity.

These systems run on private infrastructure. Below is a walkthrough of how they work in practice.

A Day with the Agents

Two agents with distinct roles — a personal assistant managing daily operations and a communications director monitoring the scientific literature.

6:30 AM — Personal Assistant sends Morning Briefing via Telegram

Good morning. Here’s your Tuesday briefing.

Today’s Schedule

  • 10:00 AM — Weekly Lab Meeting
  • 1:00 PM — T32 Steering Committee
  • 3:30 PM — Office Hours

Meeting Context — Lab Meeting

From last week’s notes and Trello board:

  • Manuscript figures still pending review (assigned to you)
  • Histological analysis of brain sections — staining protocol finalized, imaging underway
  • Operant behavior analysis needs updated parameters for the ongoing feeding study
  • Trello @mention: Review experimental design plans for the new cohort

Tasks

  • Overdue: Write recommendation letter (was due Monday)
  • Due today: Finalize lab meeting agenda, Review experimental design plans
  • Tomorrow: Submit travel reimbursement

Notes

The recommendation letter is overdue — I’d prioritize that before the meeting. The lab meeting agenda looks heavy; you may want to defer the operant behavior discussion to next week to keep it under an hour.

9:15 AM — Communications Director status report

Weekly Digest complete — 150 publications evaluated and scored, 10 identified as high matches. Pushed and available for review at Research Digest.

10:45 AM — Communications Director forwards meeting request

Dr. Smith is requesting a meeting Thursday afternoon to discuss the upcoming collaboration. You have the T32 Steering Committee from 1–2, but are otherwise free. Would you like to meet at 2:00 or 3:00?

Let’s make it 3 in case the committee meeting runs long.

calendar — Creating meeting: Thursday 3:00 PM with Dr. Smith
email — Sending calendar invite to both parties

Done. Meeting scheduled for Thursday at 3:00 PM. Calendar invite sent to you both.

11:15 AM — Voice message to Personal Assistant after lab meeting

Lab meeting minutes saved to vault. Please review and extract action items for me.

read — Reading lab meeting minutes from Obsidian vault
tasks — Extracting action items and assigning priorities

I’ve reviewed the lab meeting notes. Three action items for you:

  • Review manuscript figures — due Friday, medium priority
  • Write recommendation letter — due tomorrow (moved up from overdue), high priority
  • Schedule grant brainstorm session — coordinate with two co-investigators, due Friday, medium priority

That looks great, go.

Tasks pushed to your task list.

Morning Briefings Calendar Management Meeting Coordination Literature Monitoring Task Extraction Trello Integration

How It’s Built

A Multi-Agent Support Team

LLM interactions historically occurred in a chat window, but OpenClaw changed the model by giving LLMs the ability to directly interact with the world.

To explore this space, I developed two AI Assistants. The first is a Personal Assistant to manage tasks, calendar, files, and my growing digital knowledge base. The second is a Communications Director to read the internet, monitor scientific publications, surface and summarize interesting information, and communicate with others on my behalf.

Personal Assistant

Local Machine · Inward Facing

lock
  • Task management
  • Daily notes & briefings
  • Calendar management
  • Reminders
  • Brainstorming
  • Memory & recall

Communications Director

Cloud Server · Outward Facing

language
  • Literature review pipeline
  • Paper scoring & ranking
  • Weekly publication digests
  • Full paper summaries
  • Meeting coordination
  • External communications

Key Design Decisions

AI-Supported Interaction

Both agents are autonomous. They can run tasks spontaneously or in response to queries or input.

Input Isolation

The Personal Assistant never touches untrusted external input. It only processes data from trusted sources and human-approved transfers.

Data Confidentiality

The Communications Director never touches personal data. It operates exclusively on public scientific literature and external communications protocols.

OpenClaw Foundation

Built on OpenClaw, an open-source agent framework that provides for agentic workflows, full local file system control, and the ability to interact with a large diversity of tools, skills, and apps.