Show HN: CIYA – Purely Deterministic AI
CIYA launches v0.02, a deterministic AI engine offering prompt storage, logic templating, and an on-prem compiled language model targeting sub-200ms token resolution.
CIYA launched v0.02 on Hacker News on August 19, 2026, positioning itself as a purely deterministic alternative to conventional large language models, per the Show HN post by author iiiiiiiiio.
The product is available via API and WebSocket integration and functions, at this stage, as a storage and logic engine for prompt management. According to the Show HN post, use cases include caching and reusing repeated prompts, templating business logic, and keeping that logic on-premise rather than exposing it to cloud providers. Users can access a guide within the application by running the command R: guide, which surfaces available tools including the ability to run applications, model custom datasets, and build independent response tables.
The headline technical claim is a compiled language model (CLM) the team calls fully deterministic. The Show HN post states the CLM is designed to resolve up to 1 million tokens on-premise in under 200ms without querying an external LLM. The company says it has developed a "clear path" for CLM integration, though the current v0.02 release is described as a precursor ahead of that full rollout.
No funding, founding date, team size, or headquarters are disclosed in the source material. The company's landing page is at iiio.co, and the Show HN post notes that prospective customers or partners can reach the team via that site or by running R: contact inside the application.
The pitch targets companies that run high volumes of repeated or templated prompts and want to avoid the latency, cost, and data-exposure risks associated with round-tripping to hosted LLM APIs. The deterministic framing is notable in a market dominated by probabilistic models, though no independent benchmarks or third-party validation of the sub-200ms claim are included in the announcement. The Show HN post had 1 point and no comments at the time of publication, and the team explicitly invited feedback and questions from the HN community.
No comments yet — start the thread.