Skip to main content
Define the state structures for the agent.

​
react_agent.state.InputState Objects

View Source Defines the input state for the agent, representing a narrower interface to the outside world. This class encapsulates the initial state and structure of incoming data provided by the user or extracted from the conversation history. It holds the raw code snippets to be translated and the identified source and target frameworks. Attributes:
  • messages - A sequence of messages tracking the conversational state.
  • source_schema_code - The raw source schema code string provided by the user.
  • source_query_code - The raw source query code string provided by the user.
  • translation_type - The scope of the translation (SCHEMA, QUERY, or BOTH).
  • source_target - The identified origin framework (e.g., DOTNET_EFCORE).
  • source_target_version - The version of the origin framework, if applicable.
  • destination_target - The identified target framework (e.g., JAVA_SPRING_DATA_MONGODB).
  • destination_target_version - The version of the target framework, if applicable.

​
InputState.messages: Annotated[Sequence[AnyMessage], add\_messages]

Messages tracking the primary execution state of the agent. Typically accumulates a pattern of:
  1. HumanMessage - user input
  2. AIMessage with .tool_calls - agent picking tool(s) to use to collect information
  3. ToolMessage(s) - the responses (or errors) from the executed tools
  4. AIMessage without .tool_calls - agent responding in unstructured format to the user
  5. HumanMessage - user responds with the next conversational turn
Steps 2-5 may repeat as needed. The add_messages annotation ensures that new messages are merged with existing ones, updating by ID to maintain an “append-only” state unless a message with the same ID is provided.

​
react_agent.state.OutputState Objects

View Source Defines the output state for the graph, representing a narrower interface to the outside world. This class encapsulates the final resulting data that is returned back to the user once the translation and evaluation process completes successfully or ends in failure. Attributes:
  • translated_schema_code - The final translated schema code in the target framework.
  • translated_query_code - The final translated query code in the target framework.
  • source_validation_schema_code - The fully runnable source schema setup for validation.
  • source_validation_harness_code - The fully runnable source query execution harness.
  • target_validation_schema_code - The fully runnable target schema setup for validation.
  • target_validation_harness_code - The fully runnable target query execution harness.
  • explanation_message - The final reasoning or evaluation summary provided by the LLM.
  • messages - The sequence of messages representing the final conversational state.

​
react_agent.state.State Objects

View Source Represents the complete internal state of the graph, extending InputState and OutputState. This class is used as the core memory object passed around between nodes in the LangGraph state machine. It holds intermediate artifacts, tool execution results, retry counters, and equivalence testing states throughout the agent’s lifecycle. Attributes:
  • is_last_step - A LangGraph managed flag indicating if the recursion limit is about to be hit.
  • source_validation_entry_type_name - The entrypoint class name for running source validation.
  • target_validation_entry_type_name - The entrypoint class name for running target validation.
  • source_query_validation_results - The JSON results generated by the source query execution.
  • target_query_validation_results - The JSON results generated by the target query execution.
  • query_equivalence_deep_diffs - Dictionary of entity-level equivalence diffs between source and target outputs.
  • schema_context - A textual summary of the database schema mapping used by the LLM.
  • translation_messages - An isolated message thread specifically for the translation/validation loop.
  • extraction_loop_count - Counter for how many times the extraction node has retried.
  • translation_loop_count - Counter for how many times the translation node has retried.

​
State.is_last_step: IsLastStep

Indicates whether the current step is the last one before the graph raises an error. This is a ‘managed’ variable, controlled by the state machine rather than user code. It is set to ‘True’ when the step count reaches recursion_limit - 1.

​
State.source_validation_entry_type_name: str | None

The entrypoint class name for running source validation, extracted from source_validation_schema_code. This name is used by the sandbox execution tool to locate the correct class to instantiate and run.

​
State.target_validation_entry_type_name: str | None

The entrypoint class name for running target validation, extracted from target_validation_schema_code. This name is used by the sandbox execution tool to locate the correct class to instantiate and run.

​
State.source_query_validation_results: QueryValidationResults | None

The raw JSON results from executing the source framework’s query harness in the sandbox. Contains entity data and raw outputs.

​
State.target_query_validation_results: QueryValidationResults | None

The raw JSON results from executing the target framework’s query harness in the sandbox. Contains entity data and raw outputs.

​
State.query_equivalence_deep_diffs: dict[str, QueryEquivalenceDeepDiff] | None

Stores the differential analysis between source and target query outputs. If the graph detects semantic drift, it stores entity-level diffs here to help the LLM debug and subsequent auto/manual evaluation.

​
State.schema_context: str

Used to hold large schema metadata extracted from the database via MCP tools. We store this outside of ‘messages’ to prevent the main chat history from blowing up context windows.

​
State.translation_messages: Annotated[Sequence[AnyMessage], add\_messages]

An isolated sub-graph message list. The LangGraph translation cycle can get very noisy with sandbox compile errors and retries. By separating this from messages, the core reasoning loop doesn’t get distracted by iterative trial-and-error logs.

​
State.extraction_loop_count: int

Track the number of times we’ve looped through the extraction node to prevent infinite recursion if the LLM fundamentally fails to parse user intent.

​
State.translation_loop_count: int

Track the number of times we’ve retried a translation after sandbox failures or deepdiff failures. The graph uses this to branch to human_intervention_node if it exceeds the limit (e.g., > 3).

​
State.ui_messages: Annotated[Sequence[AnyMessage], add\_messages]

A separate message list intended for UI-only display purposes.