Skip to main content
Adapt-1 gives applications direct access to Core. Start with the result your application needs, not with an architecture diagram or learner configuration. Most first integrations choose among three paths:
  • Use a stateless route when one input should produce one result and nothing from that call should change later calls.
  • Use the State gateway when prior evidence, constraints, or decisions should remain available, but the workflow does not need a declared task grammar or adaptive outcome contract.
  • Use a Domain when the application needs a stable task vocabulary, event and outcome meanings, a contract-defined result, or task-scoped adaptation. Start with Discovery when the task boundary is known but you do not want to author the learner-facing ontology in advance.
After choosing the state boundary, fix the complete learning-run boundary. Decide whether task-specific state starts empty or was earned in a separate acquisition phase, and whether the run may update it. See Choose a learning setup.
A Domain is the stable public contract for a task. It is not a prompt, a dataset, a solved policy, or the current episode.
1

Write the result contract

State what the application needs back: a selected action, numeric prediction, coordinates, ranked alternatives, a supported hypothesis, or another machine-readable result.
2

Choose the state boundary

Use a stateless route for independent calls. Use the State gateway for evidence and constraints without a declared task grammar. Use a Domain when public task meanings, choices, outcomes, or adaptive relationships must remain scoped together.
3

Fix the learning-run boundary

Declare the complete interval used for the learning or evaluation claim. Start empty when that run should earn its own task-specific state. Use a separate acquisition phase when the run must begin from learned state formed on excluded records. Then decide whether learner writes are allowed during the run. Episode and pass boundaries do not create a new warm start when state was earned earlier inside the same run.
4

Describe one complete interaction

For a stateful workflow, write the sequence in ordinary language:
The application owns perception, execution, and validation. Core owns the reasoning and revisable state exposed through the selected contract.
5

Choose Discovery or authored ontology

For a Domain, define the public boundary first: event meaning, target or result, available choices when applicable, outcome semantics, and episode or safety boundaries. Use Discovery when you want Adapt-1 to form learner-facing input and representation structure from admitted evidence. Author the ontology explicitly when that structure must be fixed or reviewed in advance. This structure-source choice does not determine whether learning happens before or during the declared run.
6

Make the first direct call

Follow the Adapt-1 quickstart for authentication and exact route schemas. Use continuity across calls for the State gateway, Discovery when learner-facing structure should form from evidence, or Author a Domain ontology when that structure should be explicit in advance.

​
The seven-line Domain worksheet

Complete this before writing a Domain body:
If the consequence line is empty, the application may need retained evidence or structured reasoning without feedback adaptation. It may also need a stateless route. Do not invent a reward only to make the task look like learning.

​
Choose the next guide

Build with an AI agent

Use a coding agent to implement and verify the selected Adapt-1 workflow in an existing codebase.

Continuity across calls

Retain useful evidence and constraints when a full Domain contract is unnecessary.

Use Discovery

Let Adapt-1 discover learner-facing structure instead of authoring the ontology up front.

Design an authored Domain ontology

Use the explicit path when task-facing structure should be fixed or reviewed in advance.

Choose a learning setup

Distinguish zero-start learning from separately acquired frozen or warm-started state.

Choose how a Domain learns

Distinguish direct feedback, transition prediction, bounded scoring, delayed credit, and typed hypotheses.

Transition learning, advanced

Configure before-state, action, observable target, support, grouping, and abstention.

API overview

See how Domains, events, queries, consequences, and state fit together.