Learning during use
With Discovery, useful input fields, combinations, temporal relationships, predictive rules, and compatible action or transition bindings can form through experience. Causal learning uses admitted intervention evidence to learn relationships among variables. A fresh instance can begin with empty task state and learn from observations and measured outcomes, without a task-specific training dataset. Your application defines the available observations, operations, outcome meanings, and required result. With learning enabled, observations and outcomes can change learned relationships, rules, models and their parameters, and policy preferences. Consolidation stabilizes selected records. Unused associations can decay, and later evidence can replace or retire learned rules. Later requests use the resulting state. Reinforcement learning with adaptive structure (RLAS) connects learning from action outcomes with adaptation of the structure used to judge later choices.Choose a workflow
A Domain defines the task contract and scopes its evidence and learned state. Within that contract, learner-facing structure can be authored or formed through Discovery. Supported configurations can keep selected fields explicit while discovering others.
Connect results to outcomes
Send the information available when a result is needed. Let your application validate and use that result, then return the observations or measured consequences required by the selected workflow. The evidence returned to Adapt-1 depends on the workflow:
Each route documents the evidence it accepts and the result it returns.
Adapt-1 supports both continued learning and frozen use of acquired state. Learning schedules explains how to choose that lifecycle.
Inspect what changed
Depending on the route, responses can expose supporting evidence, competing hypotheses, conflicts, missing information, policy scores, discovered structure, and learner state. Use these alongside the application’s observed outcomes to understand and evaluate adaptation. What a trace exposes explains how to interpret the internal mechanisms, learner versions, and attribution visible in these diagnostics. Mechanisms gives more detail on the learned structure and parameters behind the results.Further reading
The Further reading page collects the technical write-ups, evaluation reports, and earlier background material. Start with adaptive state and Preview, then follow the behavioral evaluations, plasticity rules, Discovery, and continuous control.Start building
Quickstart
Authenticate and make your first Adapt-1 request.
First learned result
Send observations and query the state they form.
Choose a workflow
Match the result you need to an interface and feedback loop.
Use cases
Find an application workflow and the inputs, results, and feedback it uses.