| title | ARC-Neuron LLMBuilder |
|---|---|
| description | Governed local AI / offline LLM builder for ARC-Neuron: GGUF promotion, Omnibinary memory, Arc-RAR rollback, dataset governance, ProtoSynth/Neural Synth, and Synth companion roadmap. |
A governed local AI build-and-memory system that can train small brains, compare them, protect the better one, archive the worse one, and preserve the evidence of why.
ARC-Neuron LLMBuilder is a reproducible local AI build lab, not a black-box chatbot wrapper. Clone it to inspect the proof path, Gate v2 promotion rules, dataset quarantine flow, ARC Language Module truth spine, Omnibinary receipt memory, Arc-RAR restore bundles, and the 3.0 → 4.0 → 5.0 → 7.0 roadmap from protected model governance into ProtoSynth / Neural Synth and a future Synth companion brain lab.
git clone https://github.com/GareBear99/ARC-Neuron-LLMBuilder.gitRead: Why Clone ARC-Neuron · Synth Roadmap · llms.txt
- README — what this is and why it exists
- Quickstart — 10-minute tour
- Examples — runnable recipes
- FAQ — 20+ searchable questions
- Architecture — the four frozen roles
- Governance Doctrine — how Gate v2 decides
- Glossary — every ARC-specific term
- Comparison — vs MLflow, W&B, Langfuse, llama.cpp
- Usage — complete command reference
- Roadmap — v1.1 → v2 milestones
- Changelog — release history
- v1.0.0-governed Release Notes — evidence dossier
- Model Card —
arc_governed_v10_wave4— current incumbent (v2.0.0-audited) - Model Card —
arc_governed_v6_conversation— v1.0.0 incumbent (superseded)
- ECOSYSTEM — the seven-repo ARC family
| Repo | Role |
|---|---|
| ARC-Core | Event / receipt / authority spine |
| arc-lucifer-cleanroom-runtime | Deterministic local operator kernel |
| arc-cognition-core | Model-growth lab |
| arc-language-module | Canonical lexical truth |
| omnibinary-runtime | Binary mirror / runtime ledger |
| Arc-RAR | Archive / rollback bundles |
| ARC-Neuron-LLMBuilder (you are here) | Governed build loop |
- Current reproducible incumbent:
arc_governed_v10_wave4/ 0.9237. - v11.3 / wave5 remains candidate/staging until Gate v2 promotion evidence is regenerated.
- The ARC Language Module currently carries the living lexical/knowledge spine.
- External open-source datasets are acquisition roadmap targets only; they are not bundled, not ingested, and must enter v2 candidate isolation first.
- Omnibinary + Arc-RAR provide the portable communication, archive, replay, and rollback layer across devices.
- SEO Indexing Playbook — GitHub topics, release wording, GitHub Pages, and external indexing checklist
- 3.0 Roadmap Integration — protected roadmap baseline
- Memory Evaluation Protocol — repeated-question continuity and doctrine-retention tests
- Transitional License Roadmap — 1.0-to-3.0 licensing path
- GitHub Discussions — ask questions, share runs
- Issues — bug reports, feature requests, gate behavior reports
- Contributing — how to contribute
- Security — responsible disclosure
- Sponsor — support the work
| What | Value |
|---|---|
| Tests | 136 collected; CI runs fast subset plus core-fix subset |
| Incumbent | arc_governed_v10_wave4 @ 0.9237 |
| Promotions on record | 9 (v1, v2, v4, v5, v6, v7, v8, v9, v10) — 4 post-audit |
| Benchmark suite | 168 tasks / 17 benchmark files |
| Omnibinary append | ~6,600 events/sec |
| Omnibinary O(1) lookup | ~8,900 lookups/sec |
| Archive bundles | 12 restorable |
| License | MIT |
Source code: github.com/GareBear99/ARC-Neuron-LLMBuilder Author: Gary Doman Ecosystem: ARC family — seven repos
Later roadmap horizons are split from the 3.0 protected dataset/model release:
- 4.0: ProtoSynth / Neural Synth projection layer using Proto-Synth_Grid_Engine.
- 5.0: full Portal-esque Synth companion mockup.
- 7.0: working Synth AI companion, AGI assistant, and buildable brain lab.
Read: docs/SYNTH_COMPANION_ROADMAP_4_5_7.md
ARC-Neuron now references ARC-StreamMemory as the visual/video memory add-on being built for all LLMs and ARC-style systems. See docs/ARC_STREAMMEMORY_ADDON.md.