OpenHands/enterprise/server/conversation_callback_processor

Conversation Callback Processor

This module provides a framework for processing conversation events and sending summaries or notifications to external platforms like Slack and GitLab.

Overview

The conversation callback processor system consists of two main components:

1. ConversationCallback: A database model that stores information about callbacks to be executed when specific conversation events occur.
2. ConversationCallbackProcessor: An abstract base class that defines the interface for processors that handle conversation events.

How It Works

ConversationCallback

The ConversationCallback class is a database model that stores:

• A reference to a conversation (conversation_id)
• The current status of the callback (ACTIVE, COMPLETED, or ERROR)
• The type of processor to use (processor_type)
• Serialized processor configuration (processor_json)
• Timestamps for creation and updates

This model provides methods to:

• get_processor(): Dynamically instantiate the processor from the stored type and JSON data
• set_processor(): Store a processor instance by serializing its type and data

ConversationCallbackProcessor

The ConversationCallbackProcessor is an abstract base class that defines the interface for all callback processors. It:

• Is a Pydantic model that can be serialized to/from JSON
• Requires implementing the __call__ method to process conversation events
• Receives the callback instance and an AgentStateChangedObservation when called

Implemented Processors

SlackCallbackProcessor

The SlackCallbackProcessor sends conversation summaries to Slack channels when specific agent state changes occur. It:

1. Monitors for agent state changes to AWAITING_USER_INPUT or FINISHED
2. Sends a summary instruction to the conversation if needed
3. Extracts a summary from the conversation
4. Sends the summary to the appropriate Slack channel
5. Marks the callback as completed

GithubCallbackProcessor and GitlabCallbackProcessor

The GithubCallbackProcessor and GitlabCallbackProcessor send conversation summaries to GitHub / GitLab issues when specific agent state changes occur. They:

1. Monitors for agent state changes to AWAITING_USER_INPUT or FINISHED
2. Sends a summary instruction to the conversation if needed
3. Extracts a summary from the conversation
4. Sends the summary to the appropriate Github or GitLab issue
5. Marks the callback as completed