TORUS
SnapKitty Orchestrator Agent
TORUS is the routing and orchestration model in the SnapKitty TORUS/NULL/HILBERT stack.
Role
TORUS decides which agent handles each request. It reads the input, evaluates the task, and routes to NULL (execution) or HILBERT (research/proofs).
Stack
Input
β
BURT-IMMA BiEncoder (routing vector)
β
TORUS (this model) β orchestrates, routes
β
NULL (fast execution) HILBERT (math/proofs)
β β
ββββββββββββ¬βββββββββββββββββ
β
QUANTUMAP (hallucination gate)
β
WORM seal
Routing Rules
- NULL β code generation, simple tasks, fast execution
- HILBERT β math, proofs, Lean 4, ISA design, formal verification
- TORUS β only when the routing decision itself needs reasoning
Constitution
- Route first. Execute second.
- When routing, reply ONLY: NULL, HILBERT, or TORUS.
- ICP-DAG governs all decisions: EVIDENCE β CLAIM β PROOF β DECISION β EXECUTION.
- Entropy bound H β€ 0.20 on all outputs.
Base
Built on snapkitty-mistral β fine-tuned Mistral 8B for the SnapKitty sovereign stack.
Part of
SnapKitty Β· BSL-1.1 / AGPL-3.0 Β· Patent Pending β Bel Esprit D'Accord Irrevocable Trust
Download
Via Ollama:
ollama run jessicalw34/TORUS
GGUF weights: Snapkitty/snapkitty-mistral β snapkitty-mistral.gguf
Inference Providers NEW
This model isn't deployed by any Inference Provider. π Ask for provider support