Glyphh Ada – Precedent-based classifier and LLM fallthrough router

Glyphh released Ada, a precedent-based classifier that returns a calibrated confidence score, the number of matching precedents, and the underlying cases behind each decision, with an LLM fallthrough when no precedent is close enough. Ada runs on customer models hosted in Glyphh's cloud, isolated per organisation, and is reachable via an MCP connector at https://api.glyphh.ai/mcp, Python and npm SDKs (pip install glyphh, npm install @glyphh-ai/glyphh), or the API at https://ada.glyphh.ai. Glyphh said Ada is included in every plan including Free and is not priced per seat, per model or per call, and that it never trains on customer data.

Glyphh Ada classifies anything you have labelled examples of — tickets, transactions, documents, events and more — returns a confidence, probability gates, and the precedents behind each answer, so every decision can be audited. Each case, with its outcome A case is typed JSON in the shape your spec gives it. Recorded with the outcome the world confirmed, it becomes a precedent, and counts from the next question. From the nearest precedents A calibrated probability for each outcome, how many precedents voted, and their receipts. When nothing is close enough, Ada says so and asks. Sharper with every graded case Each graded case is a new precedent, and Ada learns its own weights from them: the parts of a case that predict the outcome count for more. Ada does not generate. It finds the precedents closest to the case, weighs them, and reports what they say. A single precedent is not consensus, and the answer says how many stand behind it. Your software asks incidents.decide { "error": { "what": { "class": "TimeoutError", "message": "payment gateway timed out", "status": 504 } } } Ada answers example { "fix": "raise circuit breaker", "confidence": 0.82, "from": 6, "last time": { "cause": "gateway rate limit hit on retry storm", "owner": "payments", "runbook": "RB-12 back off retries, raise circuit breaker" } } Your models, in Glyphh’s cloud Your models live and run in Glyphh’s cloud, isolated per organisation. Nothing to install or run. We never train on your data. The Glyphh connector Add https://api.glyphh.ai/mcp as a connector, sign in and click Allow. The glyphh plugin’s Ada skill builds the model with you and asks it. The SDKs, or the API pip install glyphh or npm install @glyphh-ai/glyphh, with an API key from the platform, calling https://ada.glyphh.ai. Each caller sees what they may A spec can put layers behind a clearance. A status bot gets the fix; an engineer also gets the cause, the owner and the runbook. Claude signs in with your Glyphh account: no key to type. Ada is in every plan, Free included, and never priced per seat, per model or per call. /plugin marketplace add glyphh-ai/glyphh /plugin install glyphh@glyphh then add the Glyphh connector, https://api.glyphh.ai/mcp, and ask Claude to build a model with you