Research
Aether AI research and technical papers
First-party technical work from Aether AI LLC (Aether Systems): verifiable AI execution, unlimited context for local models, and cryptographic commitment infrastructure. Each entry carries a status label — shipped work links to installable code, research is labeled research.
Unlimited Context
Status: Shipped — open source
The aether-context engine: billion-token reach for any local LLM (Ollama, llama.cpp, Hugging Face) via encode-and-page rather than compress-and-forget. Installable today: pip install aether-context.
Protocol-C: Closing the Instruction-to-Execution Gap
Status: Shipped — open source
A free, auditable authentication layer for AI decisions: sign each decision with a one-shot secp256k1 key, verify before execution, keep a tamper-evident audit log. Classical CSPRNG with a temporal safety margin; the quantum variant is Protocol-L.
Aether Atlas
Status: Research
Aether's mapping of model routing and constraint architecture across the platform.
Aether Actions: Hosted Execution and Predator Security
Status: Shipped — production
One execution plane from repository event to verified result: hosted Build & Test, GitHub-triggered automation gated on explicit per-repository spend ceilings, and Predator Security verifying exact commits across five locked assurance routes. Includes the production billing-conservation proof — 770,000 UVT reserved, 254,400 captured, 515,600 released, 0 unaccounted.
Predator: two distinct lanes
Predator Live / Predator-L concerns authorized live engagement: reconnaissance, security analysis, TTP-oriented reasoning and scoped workflows. It is not permission for unbounded autonomous intrusion.
Predator / Crucible research concerns code and security assurance, vulnerability/bug separation, program repair and verified classical/quantum comparisons. AQRC and Observatory research support investigation and evidence collection, not a blanket performance claim.
Quantum/classical feedback research
The research question is whether quantum-guided deep feedback can outperform matched classical deep feedback in bounded, verified program-repair and security experiments. Native verification, matched arms, reproducibility and prospectively specified experiments are the evidence standard.
Generation 1 exposed experimental and runtime weaknesses; it was not evidence of quantum computational advantage. Generation 2 is a prospectively registered experiment family, not a published advantage result. This page does not assert a live experiment status or a successful outcome. Historical quantum-to-classical-to-quantum architecture is lineage, not proof of an advantage.
Related
Commitment-chain verification and Quantum-Constrained AI research are related to AetherCloud. Definitions for every term live in the glossary; company background is on the About page.