graph TD %%{init: {'theme': 'mc','layout': 'elk'}}%% ParseData-ptp7g[<div><img src="/_astro/braces.Djq0PW4_.svg" style="height: 20px !important;width: 20px !important"/></div>Obtener Datos] style ParseData-ptp7g stroke:#a170ff Prompt-7qn0s[<div><img src="/_astro/square-terminal.BMOXc-nZ.svg" style="height: 20px !important;width: 20px !important"/></div>Consulta] style Prompt-7qn0s stroke:#a170ff LanggraphReactAgent-1wsn2[Agent] style LanggraphReactAgent-1wsn2 stroke:#a170ff OpenAIModel-kx9h5[<div><img src="/_astro/openAI.BhmuxEs3.svg" style="height: 20px !important;width: 20px !important"/></div>OpenAI] style OpenAIModel-kx9h5 stroke:#a170ff ChatOutput-ugeve[<div><img src="/_astro/messages-square.BaSDmT6g.svg" style="height: 20px !important;width: 20px !important"/></div>Chat Output] style ChatOutput-ugeve stroke:#a170ff GmailMesageComponent-riqjp[<div><img src="/_astro/gmail.CLxyIGOu.svg" style="height: 20px !important;width: 20px !important"/></div>Obtener Mensajes] style GmailMesageComponent-riqjp stroke:#a170ff GmailMesageComponent-1gc2u[<div><img src="/_astro/gmail.CLxyIGOu.svg" style="height: 20px !important;width: 20px !important"/></div>Marcar como leído] style GmailMesageComponent-1gc2u stroke:#a170ff TextInput-1cb9o[<div><img src="/_astro/type.Dy26vmDy.svg" style="height: 20px !important;width: 20px !important"/></div>Tema a comprobar] style TextInput-1cb9o stroke:#a170ff Prompt-7qn0s -.- LanggraphReactAgent-1wsn2 linkStyle 0 stroke:#a170ff OpenAIModel-kx9h5 -.- LanggraphReactAgent-1wsn2 linkStyle 1 stroke:#a170ff LanggraphReactAgent-1wsn2 -.- ChatOutput-ugeve linkStyle 2 stroke:#a170ff GmailMesageComponent-riqjp -.- ParseData-ptp7g linkStyle 3 stroke:#a170ff GmailMesageComponent-1gc2u -.- LanggraphReactAgent-1wsn2 linkStyle 4 stroke:#a170ff ParseData-ptp7g -.- Prompt-7qn0s linkStyle 5 stroke:#a170ff TextInput-1cb9o -.- Prompt-7qn0s linkStyle 6 stroke:#a170ff

📄 Unread Messages Processor

🧩 Overview

The workflow automates the detection and marking of Gmail messages as read based on a user‑defined topic. It pulls recent emails that match a specified query, formats the email content for the assistant, and then lets a language‑model‑powered agent decide which messages to mark. The process is fully automated and requires no manual intervention beyond providing the topic.

⚙️ Main Features

Topic‑driven email retrieval – fetches recent Gmail messages filtered by a search query.
Data formatting – transforms raw email data into a readable text list for the assistant.
Prompt construction – builds a dynamic instruction that includes the email list and topic.
LLM reasoning – a pre‑built Langgraph React Agent interprets the prompt and determines which messages to mark.
Gmail action tool – the agent can invoke the Gmail tool to update message labels (mark as read).
Chat‑style output – the final decision is displayed as a chat message in the playground.

🔄 Workflow Steps

Component Name	Role in the Workflow	Key Inputs	Key Outputs
Text Input	Provides the topic that the assistant should look for in the inbox.	Topic text (e.g., “LinkedIn”)	Text string “LinkedIn”
Gmail Message Component (Get Messages)	Retrieves recent emails from Gmail that match the query and label.	Gmail credentials, query string, number of emails to read	Raw email data (Data)
Parse Data	Converts the raw email data into a plain‑text list using a template.	Raw email data	Formatted email list (Message)
Prompt	Builds a dynamic instruction that contains the email list and the topic.	Email list, topic	Prompt message (Message)
OpenAI Model	Supplies the language model that the agent will use.	LLM configuration (model name, credentials)	Language model instance
Langgraph React Agent	Orchestrates the conversation, interprets the prompt, and calls the Gmail tool when needed.	Prompt message, LLM instance, Gmail tool (Mark as Read)	Agent response (Message)
Chat Output	Presents the agent’s final answer in a chat‑style interface.	Agent response	Chat message displayed to the user

🧠 Notes

The workflow relies on Gmail API credentials and OpenAI API credentials that must be provisioned beforehand.
Only the latest 20 emails (or the number specified) are retrieved; older messages are ignored unless the query changes.
The agent is limited to the “Mark as Read” tool; it cannot modify other labels or delete messages.
The system handles up to a maximum of 10 iterations per agent run, preventing runaway loops.
The prompt template is static; any changes to the instruction format require updating the Prompt component.
Error handling is built into the agent’s fallback model, ensuring graceful degradation if the primary model fails.
This workflow is intended for use in a playground environment where chat outputs can be visualised; it can be adapted to other interfaces by replacing the Chat Output component.