Choose a workflow

Choose an integration by the result you need, the evidence you have, and what should persist.

Choose the API workflow for the result your application needs. If you are deciding what to build, start with application ideas for concrete workflows and the role Adapt-1 could play in them.

New to the API? The quickstart checks authentication. First learned result contains a complete observation-to-prediction program.

What should Adapt-1 return?

I need to…	Start with	The application supplies
Predict an observable output	First learned result	Completed observations and a query with the target withheld
Learn which choice works in a context	Learn from an outcome	Decision-time context, the executed policy, and its measured outcome
Keep and correct evidence across calls	Continuity across calls	Useful records, constraints, and corrections
Discover predictive fields and relationships	Use Discovery	Eligible observations, the target, and a stable task boundary
Use later outcomes to revise earlier actions	Sequential Discovery	Ordered transitions, episode IDs, rewards, and terminal status
Acquire or refine numeric command sequences	Machina configuration	Numeric state, goal, action and timing conventions, plus an executor
Return the documented interval/scalar policy output	Adaptive policy contract	Ordered observations matching that policy’s contract
Request a spatial judgment or grounded point	Spatial Vision	Inputs accepted by the spatial route

A product can combine several of these patterns. The table helps you choose a first integration; it is not a list of product categories Adapt-1 is limited to.

Choose the scope

State gateway

Use prior evidence and constraints without defining a task grammar. Hosted identity follows the bearer token, not the body session placeholder.

Domain

Keep a task contract, its observations, outcomes, and learned state together under a domain_id.

For independent results, use an applicable route with the documented non-retaining behavior. Setting allow_exploration: false is not a general read-only switch. See HTTP and contracts.

Define one useful interaction

1. Name the result

   Write what code needs back: a numeric prediction, a selected action, a command sequence, a hypothesis, or an evidence packet. Use the selected guide’s result fields rather than asking every route to return the same shape.

2. Provide the public task contract

   For a Domain, define the observation and outcome meanings, result type, available choices when needed, and boundaries. Use Discovery when learner-facing structure should form from evidence. Use authored structure when it must be explicit or reviewed.

3. Complete the return path

   Use or execute the result, measure what followed, and send only the eligible observation or outcome. A prediction task can learn from completed records without an action or reward. A control task must preserve what actually executed.

4. Check the later result

   Inspect admission, query again, and record whether the intended retained state influenced the result. A successful HTTP write alone does not prove that learning was applied.

Keep three choices separate

Choice	Question	More detail
State scope	Which records and task meanings should stay together?	Domain design
Learning relationship	Am I learning an observable target, an immediate action outcome, or delayed credit?	Domain learning patterns
Learning schedule	Does state start empty or come from prior acquisition, and are writes allowed now?	Choose a learning setup

Discovery versus authored structure is another configuration choice; it does not determine when learning happens. A Discovery Domain can learn before frozen use, and an authored Domain can begin learning from zero during use.

Copy the Domain worksheet

Domain scope:
Public vocabulary:
Observed before the result:
Available choices:
Result Adapt-1 should return:
Observable consequence or feedback:
Episode, reset, and evaluation boundary:

A Domain declares the task interface, not its solution. Keep the correct action, hidden mapping, and withheld target out of the pre-result inputs. If the consequence line is empty, you may need retained evidence or an independent result rather than feedback adaptation; do not invent a reward to fill the worksheet.

Application ideas

Explore routing, troubleshooting, monitoring, personalization, and control workflows.

Build with an AI agent

Turn the selected workflow into a bounded implementation and test plan.