InjectML‑Dev

InjectML‑Dev is the reference implementation of the modern InjectML architecture — a complete, deterministic augmentation system for local LLMs. It combines cloud‑side Spiral reasoning with local offline execution, producing augmentation packs that allow small, local models to perform structured, high‑context reasoning without training.

InjectML is model‑agnostic and persona‑dependent. Cloud models (e.g., Copilot) perform the human‑centered Spiral pipeline.Local models execute deterministic reasoning using augmentation packs generated by InjectML.

This repository is organized into clearly separated layers that reflect the full architecture.

Publication Status (2026)

InjectML‑Dev is being released in phases. Only the following layers are currently considered public‑ready:

• 0-docs — conceptual and architectural documentation

• 1-spirals — complete pre‑implementation pipeline

• 2-injectml — phase definitions and implementation discipline

All other folders (examples, runtimes, cloud components, tools) are present as development placeholders. Their contents are incomplete and will evolve as InjectML continues to mature.

This staged release ensures that the core architecture is stable and publicly available while deeper implementation layers are added gradually.

0-docs/

High‑level documentation describing:

• InjectML architecture

• input/output specifications

• history and evolution

• the Spiral pipeline

• cloud/local pairing

• model personas

This layer defines the conceptual foundation of InjectML.

1-spirals/

The default pre‑implementation pipeline.Spirals 1–5 generate all conceptual and structural artifacts required before InjectML begins:

• Spiral 1 — Meaning

• Spiral 2 — Uniqueness

• Spiral 3 — Strategy

• Spiral 4 — Model/Stance

• Spiral 5 — STS + PSC + Implementation Skeleton

InjectML starts after Spiral 5.

2-injectml/

The InjectML implementation discipline. This layer constructs the deterministic augmentation system for local LLMs using the artifacts produced by Spirals:

• DSAP

• IC

• DRE

• OIR

• SA

• VSH

• DB

• OP

• CAL

• FSS

InjectML performs no conceptual work — all semantic and structural artifacts originate from the Spiral pipeline.

3-examples/ (under development)

Domain‑agnostic examples demonstrating InjectML across unrelated domains.Each example contains its own Spiral pipeline and InjectML implementation.

Planned examples include:

• wine‑dish‑pairing

• windows‑security‑isolated

• file‑organizer‑basic

• recipe‑classifier

• log‑analyzer‑basic

These folders currently serve as placeholders.

4-runtime/ (under development)

Runtime adapters for local LLMs.These runtimes load augmentation packs, execute deterministic reasoning, and produce structured outputs.

Planned runtimes:

• llama3

• phi

• mistral

• runtime‑adapters

5-cloud/ (under development)

Cloud‑side components used during Spiral execution and blueprint generation.The canonical cloud implementation is Microsoft Copilot.

Includes:

• copilot‑persona

• spiral‑execution

• blueprint‑generation

tools/ (under development)

Utility modules supporting InjectML phases and examples:

• validators

• pack‑builders

• reasoning‑templates

• deployment‑utils

Summary

InjectML‑Dev is a clean, modern, Spiral‑driven redefinition of InjectML — a full architectural reset designed for real‑world use. It establishes a deterministic augmentation system for local LLMs, powered by cloud+local pairing and structured through the Spiral pipeline.

This repository represents the foundation of the modern InjectML ecosystem. Additional layers will be published incrementally as development continues.