AStarship — The A* Startup
CAM, AI systems, and independent research.
AStarship builds the Automaton SCII Chinese Room Abstract Stack (Crabs) machine, tools for local model inference, and environments for learning agents. Founded and engineered by Cale McCollough in Eugene, Oregon.
Meet Cale McColloughStart here / Technical review
Get to know the work.
Read the architecture, inspect the source, and talk with the person building it. The repositories document the implementation and current development status; this site is an introduction, not a benchmark report.
01 / Architecture
Start with ASCII Crabs
The stack machine, SCRIPT specification, data layout, and testing approach.
Explore the core
02 / Inference tooling
Inspect LLMLoad
Local model launching and throughput testing with llama.cpp on mixed GPU hardware.
Read about the tooling
03 / The founder
Meet Cale McCollough
My project history, electronics work, public profile, and a direct way to get in touch.
About & contact
01 / Research
A new theory of spacetime.
Gradient Relativity is my proposed framework for relating gravity and electromagnetism through a geometric model of condensed space. The paper and introductory video are public for examination and discussion.
02 / Mission Control
Public repositories.
Explore the software stack, AI environments, and inference tools. Each project page links to its public repository. Some components are still in development; check the source before evaluating a deployment.
ASCIICrabs
The Chinese Room Abstract Stack (Crabs) Machine: a C++ implementation of the SCRIPT Specification with contiguous data structures, parsing, integer formatting, and seam-tree testing.
CrabsTK
A C++ toolkit project for connected software and art, including application APIs, Script2, and prototype IMUL language work.
CrabsBed
An embedded-C++ firmware project with controller source and a virtual breadboard for prototype tests. Raspberry Pi and BeagleBone-AI are hardware targets.
SubsecondDb
A hot-to-archive database project: Postgres scaffold, GPU scheduler source, and a roadmap for separate KV/auth and vector-search tiers, keyed by SubsecondId.
SubsecondId
Time-ordered 64-bit and 128-bit identifiers with timestamp, ticker, and source fields for hot-to-archive databases and distributed-system indexing.
iGeek
A C++ gym and world-simulation project for learning agents. Portable robotics, games, and scientific computing are its development goals.
LLMLoad
A llama.cpp launcher and token-throughput load tester that makes it easy to run multiple models on mixed GPUs — Intel Arc, ROCm, CUDA — with one command.
ProppaTP
A CPU-stage prototype exploring prompt preprocessing and tensor-parallel decoding. GPU execution, nvFP4 loading, and split prefill/decode are roadmap targets.
03 / Contact
Talk engineering with the founder.
Interested in a technical partnership or research collaboration? Meet Cale, look through the project history, and start a conversation through the public profile linked on the About page.
