The workflow begins with a trigger component that can be a Chat Input for manual testing, a Webhook for event-driven processing, or an external scheduler. The system retrieves Gmail messages using the Composio Gmail component configured as an Agent tool to search, list messages, or fetch entire threads. This component handles OAuth authentication and provides access to Gmail's functionality through a simple interface.

The processing stage uses multiple components working together. A Prompt Template supplies classification instructions and defines categories or JSON schemas for the AI model to follow. A Language Model or Agent performs the actual email summarization and classification tasks, while Structured Output components enforce consistent JSON formatting for the results. The system can optionally embed email content into vector stores like Chroma, Qdrant, Pinecone, or PGVector to enable semantic search capabilities.

The output stage handles both user feedback and automated actions. Results can be displayed through Chat Output components or trigger Gmail actions like modifying labels, creating drafts, or sending responses. Gmail actions are executed through OAuth-backed connectors with appropriate scopes like gmail.readonly, gmail.modify, and gmail.labels. The system can also push results to external HTTP endpoints, spreadsheets, or databases for further processing.

The flow can be extended significantly using other Langflow components. Additional integrations through Composio enable Slack notifications, Google Sheets logging, and task creation in Notion or Asana. Vector stores like Chroma can index email content to power semantic search and retrieval-augmented generation. API Request components can handle direct Gmail REST calls for advanced functionality not exposed through standard tools, while scheduling integrations with services like Zapier can automate regular processing cycles.