💬
SMS assistant · iOS beta
Pingr
A personal assistant that lives in your text messages. Text it what you ate and it tracks calories; text it a to-do and it keeps your list, with scheduled check-ins tuned to your goals.
Expo / React Native
Python API on Fly.io
Twilio SMS
LLM parsing
- Carrier-compliant SMS: got an A2P 10DLC brand and campaign approved, with an opt-in flow that verifies numbers by one-time passcode and handles STOP/HELP.
- Tiered delivery: per-plan daily text budgets enforced server-side, with unused check-ins falling back to push notifications.
- Shipped as a company: runs under Attis Labs LLC, through App Store review without a rejection.
📓
Desktop AI notes · in development
Spine
An AI that lives on your desktop instead of in a chat window. Answers land directly in your notes, and the notes are plain Markdown files you own, not rows in someone else's database.
Python
Local models via Ollama
pgvector
Android (S Pen)
- Filesystem-first: real
.md files are the source of truth, so notes work with any editor and never get locked in.
- Desktop overlay: chat, notes, and to-dos detach into floating panels around your work, then merge back into one window.
- Local or cloud AI: designed to run on local models or hosted ones, with retrieval over your own notes.
🌐
This site · live
attislabs.com
The company site and waitlist backend, running at the edge on Cloudflare.
Cloudflare Workers
D1 (SQLite)
Vanilla HTML / CSS / JS
- Waitlist API: validation, honeypot bot filtering, CORS allowlist, and per-IP rate limiting using salted hashes so no raw IPs are stored.
- Video streaming: the Worker answers HTTP Range requests with 206 partial content so demo videos seek and play reliably.
- Attribution: signups record their source, with a key-protected stats endpoint that reports counts without exposing any emails.
⌨️
Windows desktop app · open source
Better Terminal
A Windows terminal organized around project folders instead of tabs. Real local shells, split panes, and command history that stays with the folder each command ran in. No accounts, cloud sync, or telemetry.
Electron
TypeScript / React
xterm.js + node-pty
Playwright
- Project-aware completion: reads
package.json, pyproject.toml, Cargo.toml and more to suggest the scripts a project actually defines, without ever running project code.
- Safe shell integration: reads the live PSReadLine/Readline buffer and re-verifies it before replacing anything, and steps aside when an interactive program needs Tab and the arrow keys.
- Measured, not guessed: startup benchmarks and smoke tests drove the switch away from a self-extracting build that was slow and hung in testing.
🎮
Local AI agent · open source
FF7 automation
A local AI agent that plays Final Fantasy VII in an emulator in the background. It reads game state straight from memory and drives a virtual controller, so you can keep using your PC while it plays.
Python
Local LLM
ViGEm virtual gamepad
DuckStation
- Read-only memory decoding: field position, story flags, party HP, and battle state come from verified RAM addresses. It never writes to game memory, sends keystrokes, or steals window focus.
- Data-driven story route: each step is tied to a game flag that confirms it, so reloading a save resyncs the bot, and anything unverified stops as "blocked" instead of guessing.
- Model where it matters: the LLM picks objectives at decision points, while a deterministic movement loop re-reads state every ~30 ms and holds input continuously instead of waiting on inference.
🎞️
Generative video · prototype
Diffusion motion sequences
Turns a single image into a short animated clip by moving it along motion vectors and lightly re-rendering each frame with Stable Diffusion.
Python
Stable Diffusion img2img
Hugging Face diffusers
MoviePy
- Low-denoise img2img: each shifted frame is refreshed at 0.2–0.3 strength, keeping detail stable from frame to frame while still letting the model fill in new content.
- Runs anywhere: uses the GPU when CUDA is available and falls back to CPU otherwise, then exports straight to MP4.