Adapt-1 API

Adapt-1 is a non-LLM learning and reasoning system, and currently the most direct API access to Core. It updates retained state from admitted observations and measured outcomes. Later requests use that state to produce structured results. Retained state includes learned representations, executable rules, predictive models and their trained parameters, policies, and acquired procedures in control workflows. Neural computation and model training are part of Core’s adaptive structure. Reinforcement learning with adaptive structure (RLAS) describes how measured outcomes teach useful decisions while the structure supporting those decisions remains adaptable. Mechanisms explains action credit, structural revision, and native model training. Applications consume the returned predictions, decisions, hypotheses, evidence, or numeric action sequences under the selected API contract.

​

What learning changes

Controlled evaluations have shown improved predictions and decisions after learning, including useful behavior on unseen cases with learner updates stopped. The comparisons vary what the system can learn or use while keeping the task and evaluation conditions fixed: outcomes admitted or withheld, learned structure enabled or disabled, and an acquired controller evaluated with or without its learned correction. The recorded examples in these docs show how to inspect the resulting state and connect it to later behavior. Learning schedules explains frozen use; Further reading collects the technical write-ups and evaluation reports.

​

Use cases

Build forecasting and classification systems, discover causal structure, optimize experiments, or learn routing and scheduling decisions. Other applications include fault diagnosis, spatial reasoning, robotic manipulation, and precision positioning. The use cases page describes each application’s inputs, returned results, and feedback. The trading, experimentation, and control walkthroughs show how observations and outcomes update their learners.

​

Match a result to an interface

Combine these workflows when their inputs, outcomes, and state scopes are compatible. Choose a workflow compares their contracts.

​

How the loop works

Completed observations update a predictor. Policy learning uses outcomes attributed to the executed action. The application validates the returned result before using it and records the consequence for the applicable learner.

​

Choose the right API surface

Each interface has its own request and response contract. Start with Core concepts for Domains, events, learning, and state scope, or HTTP and contracts for authentication and route behavior. Read Core concepts for terminology, or Mechanisms for how Adapt-1 learns and uses retained state.