Sovereign AI infrastructure. WORM-sealed agents, formally verified kernels, and models you can run on a RTX 3080.
Models, agents, Lean 4 proofs, CUDA kernels, and WORM chains — built for deterministic execution. Every decision is cryptographically sealed. Every proof compiles with zero axiom admits.
Browser JIT LLM agent (Llama 3.2 1B) with Three.js 3D engine. No server needed.
Launch DemoEmojiScript bytecode, SoulVM debugger, LTMS truth maintenance — all in browser.
Launch Demo79 theorems, interactive SovMonster proof suite. Lean 4 formalization live.
Launch Demopip install summon
python -m summon.finetune --model sovereign-mimo-4b
ollama create sovereign-mimo-4b -f Modelfile
ollama run sovereign-mimo-4b
| Model | Params | What it does | VRAM |
|---|---|---|---|
| sovereign-mimo-4b | 4B | Code reward model. FSM + ERE gates + WORM seal. | 3.3 GB |
| sovereign-qra | — | Deterministic routing tensor. Zero entropy. Lean 4 proof. | — |
| snapkitty-merged | 4.2B | Nemotron Mini GGUF (Q4_K_M). Sovereign fine-tune. | 2.6 GB |
| hilbert | 4B | CUDA kernels: RMSNorm, FlashAttn, SwiGLU, RoPE. | — |
Small, local-first language models. Nemotron fine-tunes, reward models, GGUF exports for Ollama.
14 reposBOB family, sovereign kernels, verified multi-agent stacks with ERE gates and WORM-sealed execution.
22 reposZero-sorry Lean 4 theorems, Agda formalizations. 30+ theorems, 79 Jacobian proofs, entropy bounds.
16 reposCUDA for Ampere, Ada bare-metal processors, Apollo Guidance Computers in 7 languages.
10 reposGrassroots builder collective. Guilds: Forge, Cipher, Herald, Prism. Open membership.
8 repos| Dataset | Records | Description |
|---|---|---|
| sovereign-training-corpus | 882 | Curated prompt/completion pairs from WORM-sealed agent execution |
| cartographer-corpus | 67 | Domain-specific training across 9 chapters |
| worm-chain-archive | 39 MB | SHA-256 sealed execution records from 11 agents |
| sovereign-papers | 25 | LaTeX research papers with compiled PDFs |
25+ LaTeX papers: PIRTM, octonions, Coxeter/Weyl, EmojiScript, entropy theorems, cryptanalysis.
publishedFormal math tournament. Nova (sovereign fine-tune) defeated Nemotron. 8 published papers.
8 papersNLBHE, E7 lattice, black hole gravity, F4 algebra. 30+ zero-sorry Lean 4 theorems.
lean4