openedx_ai_extensions.workflows.orchestrators package

Contents

openedx_ai_extensions.workflows.orchestrators package#

Submodules#

openedx_ai_extensions.workflows.orchestrators.base_orchestrator module#

Base orchestrator class for AI workflow execution.

class openedx_ai_extensions.workflows.orchestrators.base_orchestrator.BaseOrchestrator(workflow, user, context)#

Bases: object

Base class for workflow orchestrators.

classmethod get_orchestrator(*, workflow, user, context)#

Resolve and instantiate an orchestrator for the given workflow.

This factory method centralizes orchestrator lookup and validation. It ensures that the resolved class exists and is a subclass of BaseOrchestrator, providing a single, consistent entry point for orchestrator creation across the codebase.

Parameters:

  • workflow – AIWorkflowScope instance that defines the workflow configuration.

  • user – User for whom the workflow is being executed.

  • context – Dictionary with runtime context (e.g. course_id, location_id).

Returns:

An instantiated orchestrator for the given workflow.

Return type:

BaseOrchestrator

Raises:

  • AttributeError – If the configured orchestrator class cannot be found.

  • TypeError – If the resolved class is not a subclass of BaseOrchestrator.

run(input_data)#

openedx_ai_extensions.workflows.orchestrators.content_suggestions_orchestrator module#

Orchestrator for course-wide content improvement suggestions.

class openedx_ai_extensions.workflows.orchestrators.content_suggestions_orchestrator.ContentSuggestionsOrchestrator(workflow, user, context)#

Bases: CourseSessionOrchestrator

Orchestrator that reviews a whole course’s actual content via OpenEdXProcessor.get_location_content (called with the course ID), then asks an LLM to propose content improvement suggestions per unit.

The full suggestion list (with section/subsection/unit ancestry) is always persisted course-wide in the session. What gets returned to a given request is filtered by the caller’s location: no location (or a location outside the course tree, e.g. the course outline page) returns everything; a specific section/subsection/unit location returns only the suggestions under that node.

clear_session(_)#

Clear generated suggestions but keep the session row (and its extra_instructions) alive, so the next author sees the same guidelines prefilled instead of starting from a blank field.

get_current_session_response(_)#

Retrieve the current session state, filtered to self.location_id.

status distinguishes “never generated / cleared” (no_suggestions) from “generated at least once” (completed, though the filtered list may legitimately be empty for this location). Either way, extra_instructions is always returned so the request form can prefill the course’s stored guidelines.

run(input_data)#

Fetch the whole course’s content and ask the LLM for content improvement suggestions, then return the subset relevant to self.location_id.

openedx_ai_extensions.workflows.orchestrators.direct_orchestrator module#

Orchestrators for handling different AI workflow patterns in Open edX.

class openedx_ai_extensions.workflows.orchestrators.direct_orchestrator.DirectLLMResponse(workflow, user, context)#

Bases: BaseOrchestrator

Orchestrator for direct LLM responses. Does a single call to an LLM and gives a response.

run(input_data)#

Executes the content fetching, LLM processing, and handles streaming or structured response return.

class openedx_ai_extensions.workflows.orchestrators.direct_orchestrator.EducatorAssistantOrchestrator(workflow, user, context)#

Bases: SessionBasedOrchestrator

Orchestrator for educator assistant workflows.

Generates quiz questions and optionally stores them in content libraries.

Two modes: - Direct mode (library_id in input_data): generate + commit immediately (legacy). - Iterative mode (no library_id): generate → store in session → review → save separately.

get_current_session_response(_)#

Retrieve the current session state.

  • If a collection was already saved: return the collection URL.

  • If questions were generated but not yet saved: return them for review.

  • Otherwise: return None.

regenerate_question(input_data)#

Refine an existing question at the given slot index.

Uses a dedicated refinement prompt so the LLM improves the provided question rather than generating an unrelated new one.

run(input_data)#

Generate quiz questions.

If library_id is present in input_data, immediately commit to library (legacy path). Otherwise store questions in session metadata for iterative review.

save(input_data)#

Commit selected questions to a content library.

Expects input_data with library_id, questions list, and optional publish flag.

openedx_ai_extensions.workflows.orchestrators.flashcards_orchestrator module#

Orchestrators for handling different AI workflow patterns in Open edX.

class openedx_ai_extensions.workflows.orchestrators.flashcards_orchestrator.FlashCardsOrchestrator(workflow, user, context)#

Bases: ScopedSessionOrchestrator

Orchestrator for flashcards generation using LLM.

Does a single call to an LLM and gives a response.

get_current_session_response(_)#

Retrieve the current session state.

  • If flashcards were generated but not yet saved: return them for review.

  • Otherwise: return None.

run(input_data)#

Executes the content fetching, LLM processing, and handles streaming or structured response return.

save(input_data)#

Saves the generated flashcards to the database or a file. This is a placeholder implementation and should be replaced with actual saving logic.

openedx_ai_extensions.workflows.orchestrators.mock_orchestrator module#

Mock orchestrator for testing and development.

class openedx_ai_extensions.workflows.orchestrators.mock_orchestrator.MockResponse(workflow, user, context)#

Bases: BaseOrchestrator

Complete mock orchestrator. Responds inmediately with a mock answer. Useful for UI testing.

run(input_data)#
class openedx_ai_extensions.workflows.orchestrators.mock_orchestrator.MockStreamResponse(workflow, user, context)#

Bases: BaseOrchestrator

Complete mock orchestrator with streaming. Responds inmediately with a mock answer in a streaming fashion. Useful for UI testing.

run(input_data)#

openedx_ai_extensions.workflows.orchestrators.session_based_orchestrator module#

Session-based orchestrator.

class openedx_ai_extensions.workflows.orchestrators.session_based_orchestrator.CourseSessionOrchestrator(workflow, user, context)#

Bases: ScopedSessionOrchestrator

ScopedSessionOrchestrator variant whose session is shared not only across locations, but across every UI slot (AIWorkflowScope row) that points at the same profile within a course.

Use this instead of ScopedSessionOrchestrator when one profile is attached to multiple scopes (e.g. a course-outline sidebar widget and an educator-tools widget) and authors should see one shared session no matter which widget they used. ScopedSessionOrchestrator itself keeps keying sessions on scope, since some workflows (e.g. flashcards) are expected to keep a separate session per widget even when they share a profile.

Looked up via filter().first() rather than get_or_create() because a scope-keyed row for this same (user, profile, course_id) may already exist from before this class was introduced — get_or_create would raise MultipleObjectsReturned once a second scope creates its own row.

class openedx_ai_extensions.workflows.orchestrators.session_based_orchestrator.ScopedSessionOrchestrator(workflow, user, context)#

Bases: SessionBasedOrchestrator

Orchestrator that follows the scope’s location specificity for sessions.

Intentionally skips SessionBasedOrchestrator.__init__ to avoid creating a location-specific session; instead creates a course-scoped session shared across locations.

run_async(input_data)#

Launch async task for scoped sessions.

Unlike the parent implementation, this does not write location_id to the session row (which has no location_id in its unique-together lookup). Instead the current location is persisted in metadata['location_id'] so the Celery task can recover it without risking an integrity-error collision with any pre-existing location-scoped session.

class openedx_ai_extensions.workflows.orchestrators.session_based_orchestrator.SessionBasedOrchestrator(workflow, user, context)#

Bases: BaseOrchestrator

Orchestrator that provides session-based LLM responses.

clear_session(_)#
get_run_status(input_data)#

Get the status of an async task from session metadata.

Returns:

Status information including task result if completed

Return type:

dict

run(input_data)#
run_async(input_data)#

Launch async task to execute the run method.

Parameters:

input_data – Input data to pass to the run method

openedx_ai_extensions.workflows.orchestrators.threaded_orchestrator module#

Orchestrators Base classes to hold the logic of execution in ai workflows

class openedx_ai_extensions.workflows.orchestrators.threaded_orchestrator.ThreadedLLMResponse(workflow, user, context)#

Bases: SessionBasedOrchestrator

Threaded orchestrator for conversational workflows.

lazy_load_chat_history(input_data)#

Load older messages for infinite scroll. Expects input_data to contain current_messages (count) from frontend. Returns only new messages not already loaded, limited by max_context_messages.

run(input_data)#

Module contents#

Orchestrators package for AI workflow execution.

This package provides the base orchestrator and concrete implementations.

class openedx_ai_extensions.workflows.orchestrators.BaseOrchestrator(workflow, user, context)#

Bases: object

Base class for workflow orchestrators.

classmethod get_orchestrator(*, workflow, user, context)#

Resolve and instantiate an orchestrator for the given workflow.

This factory method centralizes orchestrator lookup and validation. It ensures that the resolved class exists and is a subclass of BaseOrchestrator, providing a single, consistent entry point for orchestrator creation across the codebase.

Parameters:

  • workflow – AIWorkflowScope instance that defines the workflow configuration.

  • user – User for whom the workflow is being executed.

  • context – Dictionary with runtime context (e.g. course_id, location_id).

Returns:

An instantiated orchestrator for the given workflow.

Return type:

BaseOrchestrator

Raises:

  • AttributeError – If the configured orchestrator class cannot be found.

  • TypeError – If the resolved class is not a subclass of BaseOrchestrator.

run(input_data)#
Contents