Glyphh AI launches Ada beta classifier built without a transformer

Glyphh AI has released Ada in beta, a natural-language classifier that uses hyperdimensional computing instead of a large language model. The Georgetown, Texas company says the system is faster, cheaper and more deterministic for repetitive business decisions, with public benchmarks and free access now available.

Why it matters: - Ada is aimed at repeat business decisions, such as loan approvals, fraud flags, routing, refunds and risk scoring. - Glyphh AI says Ada can make those calls deterministically, with calibrated probabilities and a record of the precedents behind each answer. - The company positions Ada as a lower-cost alternative to using a frontier model for decisions that have already been made many times before.

What happened: - Glyphh AI launched Ada in beta on Oct. 30, 2026. - The Georgetown, Texas company describes Ada as a classifier built on hyperdimensional computing rather than a large language model. - Ada is free to start at platform.glyphh.ai. - Documentation is available at docs.glyphh.ai, benchmarks are at github.com/glyphh-ai/glyphh, and the launch post is at Glyphh's announcement.

The details: - Ada reads natural language in records such as tickets, notes and claim descriptions without a language model in the loop. - Glyphh AI says the system runs in milliseconds and keeps text inside the organization. - Ada uses a business's own labeled records to classify new ones based on what they most resemble. - The system compares typed fields and text together. - Ada learns which fields and words matter from the customer’s own data. - Ada can produce categories, ordered scores, yes-or-no outputs with probabilities, amounts, multiple labels with separate probabilities, and conditional questions. - Glyphh AI says Ada does not require retraining when a new labeled record is added. - The company says the same input always produces the same answer. - Ada returns calibrated probabilities and the cost of each choice instead of a similarity score. - When Ada has no close precedent, it can hand the decision to a person or a language model. - Glyphh AI says that handoff becomes a new precedent for later decisions. - The company says Ada takes about 6 milliseconds per decision and costs a tiny fraction of a cent. - Models run isolated per organization. - Glyphh says it does not train on customer data.

Between the lines: - Ada is built to sit in front of a language model, not replace one. - The pitch reflects a growing split in AI use: deterministic systems for routine decisions, and LLMs for cases that need broader reasoning. - In Glyphh's published bank-message routing test, Ada handled roughly four in five messages itself and sent the rest to Claude. - Glyphh says the combined system beat the model alone on accuracy at about one-thousandth of the cost. - The company says every result from that test is public on GitHub. - The product name references Ada Lovelace, credited with the first published program for Charles Babbage's Analytical Engine.

What's next: - Ada is available now in beta, with free access for early users. - Glyphh is pointing users to its docs and GitHub benchmarks as the main way to evaluate the product. - The company appears to be betting that explainable, low-cost decisioning will win adoption where businesses already have a strong history of labeled outcomes.

The bottom line: - Glyphh AI is trying to make AI useful for the narrow, high-volume decisions companies repeat every day, without calling a transformer each time.