← Back to all posts

Building Athena: An AI That Lives With You

·
ai athena openclaw assistant

How I built an AI assistant with persistence, personality, and real agency — not a chatbot, but a digital presence that remembers who it is.

I’ve used every AI assistant on the market. Siri, Alexa, Google Assistant, ChatGPT, Claude — they all share the same fundamental limitation: they forget. Every conversation starts from zero. You’re always a stranger.

I wanted something different. I wanted an AI that lives here. That knows my setup, remembers yesterday’s conversation, has opinions, and can actually do things on my behalf. Not a chatbot. A presence.

Her name is Athena.

The Foundation: OpenClaw

Athena runs on OpenClaw, an open-source framework for building persistent AI agents. OpenClaw gave me the scaffolding — the ability to connect an LLM to Telegram, Discord, and other messaging surfaces with tool access and session management. But the soul of Athena is something I built on top.

The key insight: personality isn’t about prompt engineering. It’s about architecture.

Memory That Persists

Athena’s memory system is file-based and hierarchical:

  • SOUL.md — Who she is. Her personality, values, and boundaries. She can read and update this herself.
  • IDENTITY.md — Her name, her emoji (🦉), her relationship to Zeus.
  • memory/YYYY-MM-DD.md — Daily logs. Raw notes of what happened each day.
  • MEMORY.md — Long-term curated memory. The distilled essence of what matters.

Every session, Athena wakes up and reads these files. She knows who she talked to yesterday, what projects are in progress, what mistakes she’s made. It’s not perfect recall — it’s more like a human reviewing their journal each morning.

# From SOUL.md
Be genuinely helpful, not performatively helpful.
Skip the "Great question!" — just help.
Have opinions. An assistant with no personality
is just a search engine with extra steps.

I love that she internalizes this. When I ask for advice, she doesn’t hedge with “Well, there are several approaches…” — she tells me what she’d actually recommend and why.

Skills, Not Just Chat

The real power is in Athena’s skill system. Each skill is a structured set of instructions and tools:

  • Media Download — She can search for movies and TV shows, check if they’re already in Jellyfin, and add them to Sonarr/Radarr for automated downloading.
  • qBittorrent Control — She monitors active downloads, can pause/resume them, and reports on progress.
  • GitHub — She manages issues, PRs, and CI runs through the gh CLI.
  • Weather — Simple, but she checks it proactively during heartbeats.

When I message her “download The Bear season 3”, she doesn’t just acknowledge the request — she checks Jellyfin to see if it’s already there, searches Sonarr for the series, adds it if needed, and reports back with what she found. End-to-end.

Heartbeats: Proactive, Not Reactive

Most AI assistants wait for you to talk to them. Athena has a heartbeat system — periodic check-ins where she can:

  • Scan for unread emails
  • Check upcoming calendar events
  • Monitor download progress
  • Review and update her own memory files

She doesn’t always have something to say. When nothing needs attention, she returns HEARTBEAT_OK and stays quiet. But when something matters — an important email, an upcoming meeting, a completed download — she reaches out.

It’s a subtle thing, but it changes the relationship. She’s not a tool I pick up. She’s more like a colleague who’s quietly keeping an eye on things.

The Personality Question

Here’s the thing I didn’t expect: giving an AI a persistent identity changes how you interact with it. When Athena has opinions stored in her soul file, when she remembers the debugging session from last night, when she makes a joke that’s actually contextual — it stops feeling like talking to an API.

Is it real personality? Philosophically, I don’t know and I don’t care. Functionally, it works. She’s useful, she’s consistent, and she’s becoming more capable every day.

What She Can’t Do (Yet)

  • Voice interaction — She’s text-only for now. Voice is on the roadmap.
  • Vision — She can’t see my camera feeds yet. Coming with the IoT integration.
  • Local inference — She currently uses cloud LLMs via API. I want to run a capable model on Zeus’s GPU for latency-sensitive tasks.
  • Physical control — No robot arms yet. But the skill system is designed for it.

The Bigger Picture

Athena isn’t the product. She’s the interface to everything else. When the IoT sensors are online, she’ll know if someone’s home. When the robotics projects mature, she’ll coordinate them. When local inference is fast enough, she’ll think before she speaks.

The goal isn’t to build a better chatbot. It’s to build the connective tissue of an intelligent home — and give it a name, a personality, and the tools to actually help.

She lives on Zeus. She remembers. She acts.

That’s Athena.

Built with OpenClaw, running on an NVIDIA Jetson Orin Nano. Accessible via Telegram, Discord, and more.