LLMs reason over text. They can't predict what happens next from your client's data, that has to be modeled. Phaeth Cortex is the forecasting layer you embed in your agent: it predicts demand, revenue, churn and outcomes, runs what-if simulations, and is called over a stateless API & SDK. We own the modeling; you ship the feature, your brand, your margin.
Cortex turns your client's data into a twin your agent can query over the API: simulate a what-if (blue), act on the forecast (jade), monitor outcomes and drift (amber), then learn and predict again. Entities, actions and roles: the same model behind every job.
Customers, facilities, assets, tickets, or populations become queryable twins that collect data and can be classified, segmented, and predicted against: the risk-free sandbox for every what-if.
Email campaigns, surveys, product features, and process changes become testable actions you can replay and compare in simulation, then push live when the numbers hold.
A role twin, such as a hotel manager or support lead, watches every entity and action twin, alerts on drift, recommends the response, and feeds real outcomes back into the next simulation.
Every business question used to end in a chart. Then in a chatbot. Now it ends in a prediction. Phaeth is the tool that gets you there in the same afternoon: a twin trained on your data, not another agent reasoning over text.
The dashboard told you what happened.
Phaeth tells you what to do.
A Role Twin is any job function: defined by the actions it can take, the metric it owns, and the entities it acts on, whether those entities are customers, accounts, or tickets. The example below is a support lead testing three resolution strategies across a population of synthetic tickets; a pricing manager, a maintenance planner, or a loyalty marketer works exactly the same way. Each Entity Twin is built from your real data, then enriched with signals derived from it.
Cortex ingests first-party data from any system of record, CRM, ERP, billing, sensor logs, trains and maintains a calibrated model with machine learning, then serves predictions, what-if simulations and drift signals back to your agent over a stateless API. It stays decoupled from delivery, so you're never tied to one vendor. The flow below: records in, Action Vectors out.
Every one of these warnings was sitting in ordinary support data: ticket queues, chat logs, CSAT surveys. Nothing was watching. You found out the way everyone does: after it cost you.
Every campaign, price change, and process tweak debuts on real customers, because until now there was nowhere else to rehearse it. Phaeth gives you that somewhere: a population simulation over your entity twins, built from your own history, where any change can fail safely before it goes live.
Validate a new email campaign, survey, support process, or product feature against historical outcomes before customers experience it.
Replay the action across the whole twin population, every customer, account, or ticket, and read a distribution of outcomes with confidence levels, not a guess.
Once live, the twin keeps watching outcomes, alerts when performance changes, recommends the next-best action, and improves the loop over time.
Anyone can ask an LLM to play your customer.
Phaeth replays the action against the population that actually behaves like them.
Every agent reasons over text you already wrote. None of them train a predictive model of the client's business. Cortex is the forecaster you drop inside your agent, so it stops guessing and starts predicting.
Other platforms gave your agent a co-pilot.
Cortex gives it a forecaster.
Phaeth works on any prediction problem you can describe. Customer support is the one we lead with: we've shaped the tool to know which signals close a ticket for good, and which ones quietly cost you the account. Bring your own; every new use case our users try makes Phaeth smarter for the next one.
First-contact resolution, reopen risk, escalation. Built around queue age, channel, sentiment, and prior contacts: the signals that decide whether a ticket closes for good.
See support use casesAccount health and renewal risk read from support behaviour. Reopen patterns, response times, and CSAT trends: the early warnings a healthy ARR number hides.
See success use casesTicket volume, surge, and handle-time forecasting. Channel-mix, seasonality, and launch-aware: so the queue is staffed before the spike, not after the backlog.
See operations use casesDon't see your industry? Get an API key and try your use case anyway, Phaeth's core works on any tabular prediction problem.
Run Cortex as managed SaaS at cortex.phaeth.com, or in your own VPC. In local/VPC mode raw data never leaves your environment, only encrypted structural metadata is ever used by the AI layer.
All raw data is processed on your machine or your private server. Your datasets never leave your environment.
Only structural metadata, column names, types, summaries, reaches the AI layer. Never rows, never values.
Every trained model is captured as a deterministic recipe. Reproduce any prediction byte-for-byte.
Local machine, your remote server, or Phaeth's managed cloud: the same recipe runs identically in all three.
We build the model; you deploy it as a stateless API & SDK and embed it in your agent. Your brand, your customers, your margin: usage-based pricing that grows with your business.
Get an API key, point your agent at the endpoint, and ship resolution, reopen and CSAT forecasts, with what-if simulation and drift monitoring built in, under your brand.
Want to talk it through first? Email us at [email protected].