SpaceKit AI is the public monorepo for Growformer research and machine-learning projects. It contains the core promote-freeze substrate, language-model experiments, reproducible evaluation tooling, and neural cellular automata research.
Repository: spacekit-xyz/spacekit-ai
| Directory | Role | Status | License |
|---|---|---|---|
growformer/ |
growformer-ledger/
growformer-llm/
growformer-nca/
git clone https://github.com/spacekit-xyz/spacekit-ai.git
cd spacekit-ai
cargo check --workspace
cargo test --workspace
The Growformer suite is substantially larger than the supporting crates. Use package-scoped commands during development:
cargo test -p growformer-ledger
cargo test -p growformer-nca
cargo test -p growformer-llm
cargo test -p growformer --lib
growformer ───────────────┐
├──> growformer-llm
growformer-ledger ────────┘
growformer-nca standalone
growformer
and growformer-ledger
remain independently usable.
growformer-llm
consumes their public APIs. growformer-nca
remains isolated until an explicit integration contract replaces duplicated experimental code.
Internal dependencies use workspace-relative paths during development. External consumers, including SpaceKit Core, should use tagged revisions or published crate versions rather than filesystem paths.
spacekit-ai/
├── Cargo.toml
├── README.md
├── CONTRIBUTING.md
├── SECURITY.md
├── growformer/
├── growformer-ledger/
├── growformer-llm/
└── growformer-nca/
Each project owns its detailed architecture, usage, and reproducibility documentation. Cross-project policy and CI live at the monorepo root.
These projects are experimental machine-learning systems. Passing tests or reported benchmark results do not establish general intelligence, adversarial robustness, model safety, privacy compliance, or suitability for autonomous high-stakes decisions.
Results must identify the source revision, dataset split, seed, configuration, hardware, metric, and comparison budget. Historical and negative results remain part of the research record and must be clearly labeled.
Do not commit:
- downloaded corpora or benchmark datasets;
- model checkpoints, trained brains, embeddings, or tokenized corpora;
- user conversations or production inference records;
- generated experiment reports or local evaluation ledgers;
- API keys, private keys, populated environment files, or certificates;
- build output, generated WASM packages, or dependency caches.
Small deterministic fixtures and inference configuration required by a published crate are allowed when their provenance and purpose are documented. Large datasets and models belong in versioned release assets or an external artifact store with checksums.
SpaceKit Core is a separate infrastructure monorepo. AI integration there must be optional and feature-gated. SpaceKit Core must consume tagged Growformer revisions rather than unpublished sibling paths.
Private applications and websites may consume SpaceKit AI releases, but they must not become dependencies of these public libraries.
See CONTRIBUTING.md for development requirements. Report suspected vulnerabilities privately according to
. Never place secrets, private datasets, or exploit details in public issues.
SECURITY.md
This monorepo contains packages under different permissive licenses. Consult
each package's Cargo.toml
and LICENSE
file. A root-level policy does not override package-level licensing.