AETERNA v0.9.37

What the company appears to be: AETERNA v0.9.37 is a self-reported research prototype built for the OpenAI 2026 Build Week hackathon. It claims to implement a local, non-LLM cognitive architecture using hyperdimensional computing (HDC), spiking neural networks (SNNs), and typed execution. The project includes a deterministic synthetic control scenario and a frozen, verifiable evidence system.

What changed: During the Build Week period (July 13–21, 2026), the author extended an existing research project using Codex and GPT-5.6. These tools were used for implementation, verification, falsification workflows, and packaging of a GUI showcase. The human owner selected hypotheses, thresholds, and final actions; the AI was not responsible for scientific decisions.

Single most important open question: Is this project a legitimate research prototype or an experimental demonstration with limited applicability beyond its specific synthetic context?

Note: This analysis is based solely on the self-reported description provided by the author. No external verification, traction data, revenue figures, or customer information are available.

The description states that AETERNA v0.9.37 is a local, non-LLM cognitive architecture designed to run without an LLM at runtime. It combines:

• Structured hyperdimensional representations (HDC)
• Recurrent spiking state (SNNs)
• Persistent goals
• Bounded program execution
• Protected-outcome evaluation

It also includes:

• A one-executable Windows GUI showing a live integrated Cognitive Mission and synthetic solar-and-battery control scenario.
• A source-verifiable frozen GatePack with artifact hashes, fixed seeds, deterministic replay, and machine-readable verdicts.

Inference: The product appears to be a research-grade system designed for controlled experimentation in cognitive architectures, not a commercial product or production-ready agent.

The description states that AETERNA makes the claim path part of its architecture — contract, primary, rerun, independent gate, and ledger. Negative results remain visible instead of being rewritten as success.

It also claims to be a "local non-LLM cognitive architecture" with HDC memory, recurrent SNN control, typed execution, and replayable evidence.

Inference: The positioning is centered on reproducibility, determinism, and scientific rigor in AI systems — particularly in contrast to overclaiming or post-hoc adjustments common in LLM-based systems. This is a research-oriented claim, not a commercial one.

The description does not state any explicit customer base or target market. It is described as a research prototype submitted for a hackathon competition.

Not evidenced: No information on who uses this system or what their needs are.

The description states that the current revision is proprietary under the AETERNA Proprietary Evaluation License v1.0, which authorizes OpenAI Build Week and Devpost judges to inspect the project solely for competition evaluation.

Inference: There is no evidence of a commercial business model or pricing structure beyond the limited license for judges. The system is not described as available for purchase or use outside of the competition context.

The system uses:

• Codex and GPT-5.6 for engineering and verification
• HDC/FHRR memory
• Recurrent spiking neural networks (SNNs)
• Typed Micro-VM
• Artifact verification via SHA-256
• Deterministic replay
• Frozen GatePack with artifact hashes

It includes:

• A Windows GUI executable
• Source verification scripts in PowerShell
• Git history and Codex session ID for traceability

Inference: The technical stack is oriented toward reproducibility, deterministic behavior, and verifiable execution. It is not a commercial-grade system but rather a research tool.

The description states that AETERNA was an existing research project before the Build Week event. The current submission only extends it with Codex/GPT-5.6 during a short window (July 13–21, 2026).

It includes:

• Certificates for specific synthetic scenarios
• Deterministic replay and artifact verification
• A GUI showcase

Not evidenced: No evidence of adoption, revenue, or usage beyond the competition submission.

The project was submitted to the OpenAI 2026 Build Week hackathon. It is not described as competing with any existing commercial systems. The author emphasizes its non-LLM nature and deterministic design as a contrast to LLM-based approaches.

Inference: AETERNA is positioned within a research or experimental space, not a competitive marketplace. No direct competitors are mentioned.

• The system is described as a "research prototype", not a production-ready agent.
• It is not open-source; the license limits use to competition judges.
• The synthetic solar simulation is deterministic and not validated against real-world physics.
• The project does not claim to be AGI or a general-purpose agent.
• The human owner made all scientific decisions, while Codex/GPT-5.6 was used for implementation and verification.

Inference: Risks include limited applicability beyond the synthetic context, lack of commercial viability, and restricted access due to proprietary licensing.

1. What are the core research questions or hypotheses that AETERNA is designed to test?
2. How does this system differ from other non-LLM cognitive architectures in terms of performance or scalability?
3. Are there any plans for open-sourcing or broader commercialization beyond the competition context?
4. What are the limitations of the deterministic synthetic environment, and how do they impact real-world applicability?
5. How is the human-AI collaboration structured — what decisions were made autonomously by Codex vs. by the human owner?