What it is
A public GitHub repo describing a multi-model AI companion architecture that ingests sensor/biometric signals and coordinates “lobes” via an asynchronous orchestrator and file-based IPC.
Gabriel’s notes
Project Aurelia is a public GitHub repository that ships a design-heavy README plus Python code and a “research paper” style markdown describing an “Asynchronous Biomimetic Multi-Model Entity (ABMME).” The core idea is to run multiple model “lobes” (an executive/conscious model, a background action model, and a vision model) while also ingesting physical-world signals (e.g., mmWave-derived heart rate/respiration, proximity LiDAR, and an accelerometer) and translating those into state that influences the system’s behavior. Licensing is Unknown / not confirmed (no LICENSE file was found at the repo root at the time I bookmarked this).
Quick take: I have not used or deeply explored this repo, so treat it as untrusted until you manually verify what it does (and what it stores). That said: conceptually, it’s a spicy attempt at “embodying” LLMs—less chat box, more weird little digital organism.
I saved this under AI because it’s a concrete example of an “LLM + sensors + agent loop” architecture that tries to give a model something like a nervous system (and a memory that isn’t just a giant pastebin).
What grabbed me:
- Multi-lobe split (executive vs. labor vs. vision) to reduce “compute starvation” and keep the interactive persona responsive.
- Sensor fusion as first-class input: heart rate / respiration (via mmWave pipeline), distance/presence (LiDAR), vibration/typing intensity (ADXL345), plus “somatic” mapping from hardware thermals.
- File-based IPC (“drop folders”) as a reliability tactic instead of fragile real-time pipes. Old-school, slightly cursed, but often effective.
- Memory design: hybrid retrieval described as ChromaDB + SQLite FTS5 with Reciprocal Rank Fusion, plus an end-of-session “dreaming” compression step.
Good fit if you want to:
- Study an “embodied agent” pattern without waiting on a vendor roadmap.
- See one approach to splitting “persona reasoning” from “background task execution.”
- Prototype a sensor-driven state machine that nudges an LLM’s behavior loop.
- Borrow ideas for robust local IPC (especially if WebSockets have hurt you before).
- Explore memory consolidation patterns beyond plain vector search.
Pricing snapshot (auto-enriched):
The repo is free to view on GitHub. Realistically, the “price” is the hardware/sensors (and the time you’ll spend turning a cool architecture doc into something you trust). Licensing terms are Unknown / not confirmed, so treat reuse as “ask permission or don’t ship it.”
Work-use / compliance snapshot (auto-enriched):
This project explicitly discusses collecting physiological signals (e.g., heart rate/respiration) and other sensor telemetry. That’s a compliance booby-trap if you use it at work: policies may treat this as biometric and/or health-adjacent data. If you’re subject to laws like Illinois’ BIPA or GDPR-style regimes, do a legal review before any real-person deployment. The repo does not clearly advertise a consent flow, retention policy, or threat model—Unknown / not confirmed.
Alternatives (auto-enriched):
- OpenBCI: better if your goal is biosensing/BCI tooling and hardware ecosystems; less focused on “LLM companion architecture” as a cohesive agent stack.
- Home Assistant: better if you want a mature, local-first automation/sensor integration platform; you’d still need to design your own LLM/agent loop on top.
Before you adopt it:
- Confirm the license situation (or lack thereof) before you copy code into anything you sell.
- Run it in a sandbox first; assume untrusted until you’ve audited dependencies, networking, and storage paths.
- Decide up front what telemetry you will not collect, and set retention limits (especially for biometrics).
Sources
- https://github.com/anitherone556-max/Project-Aurelia
- https://github.com/anitherone556-max/Project-Aurelia/blob/main/Project_Aurelia_Research_Paper.md
- https://gdpr-library.com/article/9
- https://www.akingump.com/a/web/101105/Biometric-Information-Privacy-Act-740-ILCS-14-1-et-seq..pdf
- https://www.home-assistant.io/