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.
When Shared State Becomes Authority
Status: Published — safety specification
A practical threat model for persistent autonomous agents, Unlimited Context, Shared IR, corpus discovery, QUBO workflows, and hybrid quantum execution. The paper separates published requirements from controls that have runtime enforcement evidence.
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.