Glyphh Ada Cuts LLM Routing Cost From $300 to $0.22 With Precedent-Based Classifier

Successful-Seesaw525 · reddit · 2026-10-24

The team behind Glyphh Ada launched a precedent-based classifier (HDC-based, not transformer) that sits in front of LLMs. Seeded with labeled historical cases, it returns a decision, confidence score, and the exact reference cases backing it; low-confidence cases fall through to an LLM or human, giving deterministic, auditable output.

Banking77 benchmark (200 support messages routed across 77 categories):

Ada alone: 90.0% accuracy
Claude (best model): 96.5% at $300 total API cost
Ada+Claude hybrid (Ada handles high-confidence, routes 22% edge cases): 97.5% at $0.22 total

Offered as a hosted API with Python/TS SDKs, an MCP plugin for tuning inside Claude, a free tier, and a no-training-on-customer-data promise. The authors note it's a single 200-message sample, with methodology, scripts, and a dataset where Ada performs poorly all public.

Original post →

Glyphh Ada Cuts LLM Routing Cost From $300 to $0.22 With Precedent-Based Classifier

More from coding & agent