Helix Models Overview
From Helix Project Wiki
🧠 Helix Models Overview
This page provides a high-level summary of all AI model architectures currently in development, testing, or production within the Helix ecosystem. Each model serves distinct roles, with varying levels of determinism, safety constraints, and runtime capabilities.
Model Index
| Model | Status | Params | Role | Notes |
|---|---|---|---|---|
| Helix LLaMA:8B | Stable | ~8B | Lightweight, deterministic responder | Ideal for constrained runtime environments. |
| Khronos | Experimental | ~175B | Meta-reasoning agent | Active under sandboxed, high-trust conditions. |
| Helix GPT – Operating Ontology | In review | TBD | Narrative, analytical assistant | Used in longform synthesis and QSR reasoning. |
| Helix Babbage (Code Agent) | Internal testing | TBD | Agent for structured code gen + audits | Part of DevOps & reproducible agent workflows. |
| Tiny Recursive Model (TRM) | Experimental | ~7M | Recursive reasoning micro-model | Excels in logic tasks, high ARC-AGI scores, ultra-low footprint. |
Model Capabilities
- All Helix models follow the Helix Core Ethos, enforcing guardrails on determinism, auditability, and safety.
- Models are aligned to specific Runbooks, which define their deployment responsibilities and human approval requirements.
- TPAF governs all irreversible operations — no model is permitted to execute actions that skip human consent checkpoints.
Runtime & Trust Zones
| Trust Level | Models | Example Use Cases |
|---|---|---|
| Production (R4, deterministic) | Helix LLaMA:8B | Audit-safe Q&A, dashboards, internal search |
| Controlled Experiment (R3-R4 hybrid) | Khronos, Tiny Recursive Model (TRM) | Recursive logic, verification agents, minimal footprint evals |
| Draft & Dev (R2+) | Helix GPT, Babbage | Exploration, ideation, soft-output agents |
