Developers
Hyper. The architecture under everything CodeNinja builds.
Hyper is the architecture CodeNinja's platforms are built on. Praxis designs a physical AI system. Hyper Ontology structures the operation's context. Hyper Pragma runs agents inside the enterprise boundary. Hyper Engram keeps the decision memory every run reads before it acts. Hyper Noesis opens the models an organization owns so their reasoning can be inspected. CodeNinja Atoms, the reference architectures on this site, is the first thing built on it. This is where developers start.
Capabilities
Five platforms on one architecture. Each does one thing in the life of an operation's context, and each is built on Hyper.
Praxis
Praxis turns an operator's requirement into a complete design for physical AI, reasoned through eight lenses from first principles to hardware, with every claim on a record and a person on every write.
Build with it: A complete system design for one operation: what to sense, where each model runs, the object model, the hardware, the three-year cost and who approves every action.
How Praxis reasonsJoin the beta
Beta, open to outside engineersHyper Ontology
Hyper Ontology holds the governed model of the organization. It imports the object model a Praxis design publishes and stands it up as a living system over the operator's own systems of record.
Build with it: A living ontology over an operator's existing systems: typed objects, typed links, and actions that sense, decide, act and learn.
How it becomes livingPackage format
Beta, a small number of teamsHyper Pragma
Hyper Pragma is agent execution inside the enterprise boundary, on models the organization can change without losing what was built.
Build with it: Agents that work on the operator's own hardware, on open-weight models the operator can swap, with the context carried across the swap.
Used in house, not yet openHyper Engram
Hyper Engram is decision memory each run reads before it acts: the record of what was decided, by whom, and what worked.
Build with it: Systems that improve run over run because every run starts by reading what earlier runs and people recorded.
Used in house, not yet openHyper Noesis
Hyper Noesis opens models the organization owns so their reasoning can be inspected, rather than taken on trust.
Build with it: Verification of a model's behaviour before it is trusted with an operation, and after every change of model.
Used in house, not yet openGetting started
Four ways in, all open today. Pick the one that matches what you are building.
Design a system on Praxis
Sign up, describe the operation, and Praxis returns a complete design. During the beta every account is approved by hand, usually within a day.
Give your coding agent every design
The MCP server lists, searches and reads the published designs: object models, model and hardware registers, cost lines and full papers.
claude mcp add codeninja-research -- uvx --from "git+https://github.com/muhammadumar89/codeninja-research#subdirectory=mcp-server" codeninja-research-mcpLoad an object model
The loader reads any published hyper-ontology/1 package, validates it, walks its typed links and converts it to Mermaid, Cypher or JSON-LD.
pip install "git+https://github.com/muhammadumar89/codeninja-research#subdirectory=hyper-ontology-py"
hyper-ontology show port-digital-twin-us
hyper-ontology cypher port-digital-twin-us > load.cypherQuery the dataset
Every design is a row in five tables: designs, objects, models, costs and full text. One load gives you all of them.
from datasets import load_dataset
objects = load_dataset("CodeNinjatools/vertical-driven-architectures", "objects", split="train")
print(objects.filter(lambda r: r["kind"] == "event")["label"])Reference
The formats and tools, documented from the repository itself.
The hyper-ontology/1 package format
What every published object model contains, the rules it follows, and what an agent does with it.
The hyper-ontology loader
Load, validate, traverse and convert packages from the command line or Python.
The MCP server
Five tools that give a coding agent every design: list, get, read, find, and where the designs come from.
The Vertical-Driven Architectures dataset
Five tables, one row per design, object, model choice or cost line; monthly DOI snapshots.
Samples
Every published design is a worked example: a complete system built on Hyper for one real operation, with its object model as a package you can load.
agriculture and earth observation
Field LedgerPakistan · 12 objects · DOI 10.5281/zenodo.23186671BaselineUnited States · 16 objects · DOI 10.5281/zenodo.23186673Fodder WatchSaudi Arabia · 14 objects · DOI 10.5281/zenodo.23186675energy and utilities
Grid Context WatchUnited States · 14 objects · DOI 10.5281/zenodo.23157957Reliability AtlasSaudi Arabia · 13 objects · DOI 10.5281/zenodo.23157965Feeder FirewatchUnited States · 14 objects · DOI 10.5281/zenodo.23159328heavy industry and construction
Structure Phase WatchSaudi Arabia · 15 objects · DOI 10.5281/zenodo.23126448Steel Count LedgerPakistan · 14 objects · DOI 10.5281/zenodo.23126563Factory Fire WatchSaudi Arabia · 14 objects · DOI 10.5281/zenodo.23126565maritime and ports
Port TwinUnited States · 13 objects · DOI 10.5281/zenodo.23126431Terminal PulseUnited States · 12 objects · DOI 10.5281/zenodo.23159331oil and gas
Sovereign HSE WatchPakistan · 12 objects · DOI 10.5281/zenodo.23119714Loop Integrity WatchPakistan · 15 objects · DOI 10.5281/zenodo.23157967Platform updates
What changed in the format, the loader, the MCP server and the dataset, from the repository's history. All updates
- Episode 6: Field Ledger (Pakistan), Baseline (United States), Fodder Watch (Saudi Arabia): agriculture and earth observation, the fifth sector
- Site address is codeatoms.ai: every canonical, package, card, dataset row, loader and MCP default moved; build artifacts untracked
- Product name: CodeNinja Atoms. Logo wordmark followed by ATOMS in every header and footer; site title, feed, llms.txt and JSON-LD renamed
- Dataset snapshot DOI: October 2026 release 10.5281/zenodo.23160820 (all versions 10.5281/zenodo.23160819)
- Version 2 of Feeder Firewatch and Terminal Pulse: Appendix A at AWS's deepest three-year plan
- Dataset card: the three episode 4 designs
- Episode 4: Grid Context Watch (US), Reliability Atlas (Saudi Arabia), Loop Integrity Watch (Pakistan)
- MCP server over every design (codeninja-research-mcp), linked from the site, llms.txt and a rebuilt README
Community
Questions, answers and what people build live in the repository's discussions. A GitHub account is all it takes.
- Announcements
What changed, release by release. - Q&A
Ask about a design, a package, the loader or the MCP server. - Show and tell
What you built on Hyper: an ontology over your systems, an agent, a tool. - Ideas
What the next design, format or tool should be.
Latest threads
- Start the first threadQ&A · open to everyone with a GitHub account
Also: the engineering blog and its feed, issues for bugs in the loader or the server, and the Hugging Face organization for every dataset and Space.