> ## Documentation Index
> Fetch the complete documentation index at: https://docs.langstate.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Orchestrator contract

> The abstract LangState lifecycle and its exact source-level method names.

`LangState` is an abstract generic class that holds component references and defines an application-facing lifecycle. The pinned source does not include a concrete subclass, so it cannot be instantiated directly.

## Required methods

An application subclass implements:

```python theme={null}
from typing import Optional

from core.langstate.base.langstate import LangState
from core.langstate.base.schema import (
    ActionResultData,
    AgentInput,
    InteractionRequest,
    LangStateConfig,
)


class ApplicationLangState(LangState):
    async def initialize(self, config: LangStateConfig) -> None:
        ...

    async def invoke(
        self,
        agent_input: Optional[AgentInput] = None,
        metadata: Optional[dict[str, object]] = None,
    ) -> InteractionRequest | ActionResultData:
        ...

    async def get_current_state(self):
        ...

    async def get_canonical_state(self):
        ...

    async def reset(self) -> InteractionRequest:
        ...
```

The interpretive accessor is `get_current_state()`. A completed invocation returns `ActionResultData`, not `StateResultData`.

## Constructor components

`LangState.__init__()` accepts optional:

* `BaseSpecExtractor`
* `BaseMutator`
* `BaseProjectorCanonicalState`
* one `BaseProjectorUI` or a list of UI projectors

The base class stores schema, canonical and interpretive state references, conversation history, and action handlers. It exposes setters and read-only component properties, but the subclass owns initialization and invocation sequencing.

## Invocation result models

`InteractionRequest` represents another user interaction. It can carry a prompt, UI components, options, pending fields, `InterpretiveStateSchema`, and `CanonicalStateSchema`.

`ActionResultData` represents completion:

```python theme={null}
ActionResultData(
    state=interpretive_schema,
    canonical_state=canonical_schema,
    success=True,
    action_data={"registration_id": "registration-17"},
    metadata={},
)
```

This model is distinct from `ActionResult` in the action module. `ActionResult` describes an individual action execution with an `ActionStatus`; `ActionResultData` is the orchestrator's completed-flow payload.

## Application responsibility

The abstract contract does not automatically connect repositories, snapshots, actions, mutators, or projectors. A subclass must define ordering, error handling, concurrency, persistence, transaction boundaries, and how component results are converted to the Pydantic boundary models.