Output Node

An Output node takes data from one upstream source and writes it somewhere. It can also (or instead) generate an AI analysis report from the data.

Output type

You choose one of three output types:

AI Data Analysis — the upstream data is summarised by the AI using a prompt you provide.
Data Export — the upstream data is written to a file or a database connection.
Both — analysis and export run in the same node.

For Data Export, you also pick a destination: a local file path or a saved connection. The downstream flow is completely different depending on which one you chose, and the chat adjusts the questions it asks accordingly.

Loading the source

Before exporting, DagFlux loads the upstream data:

Vector source — pulled in bulk from the vector index, including ids, vectors, metadata, and namespace. LanceDB sources are read through the local library; Pinecone (and other cloud vector stores) are read through their connector's bulk-fetch operation.
Local SQL or document source — read directly from the local workspace or local document file.
Remote database source — fetched through the source connector.

Local file export

When the destination is a local file, you choose the format. DagFlux supports eight formats:

JSON — plain JSON array.
CSV — comma-separated values.
Excel — single-sheet workbook.
NeDB — local document database file you can re-open later as a source.
SQLite — infers column types, creates table, inserts rows in a transaction.
Parquet — columnar format with a schema inferred from the data.
Avro — record format with a schema inferred from the data.
LanceDB — vector format, requires the source data to already contain vector fields.

The chat opens the native save dialog with a sensible default filename (based on the node id and chosen extension), and validates that you actually picked a path before proceeding.

Database export

When the destination is a database connection:

Vector destination — DagFlux delegates to the Vector Database Executor skill, which embeds the chosen column and upserts the vectors with the rest of the row stored as metadata. For Pinecone, the table name field accepts both index-name and index-name/namespace formats.
Cloud storage destination — opens the cloud storage browser so you can pick the bucket and path interactively. DagFlux serialises the data to JSON or CSV and uploads.
Document destination — DagFlux bulk-inserts the documents into the chosen collection.
Relational destination — DagFlux infers column types from the data, creates the target table if it doesn't exist, and inserts the rows in chunks. Identifier quoting is applied automatically per dialect.

Email notification

You can optionally configure the Output node to send an email when the workflow completes, with the processed data attached as a CSV. The chat captures the recipient, subject, message body, and whether to include the data attachment, plus a "send test email" button so you can confirm delivery works before relying on it in production.

AI analysis

When AI Data Analysis is selected (alone or alongside export), you write a prompt — for example, "Analyse the sales trends in this data and provide insights on seasonal patterns, top performing products, and recommendations." — and DagFlux runs the prompt against the loaded data using your configured AI provider, returning the report.

Retry behaviour

If the output fails at execution time, the chat lets you retry from the beginning without losing the conversation. Output settings are reset and reinitialised so you can adjust the destination or format and try again immediately.