Interactive Progress View

DocETL can show a full-screen, live progress dashboard in your terminal while a pipeline runs. It lets you watch each document complete, see cost and timing per operation, and click into any finished document to inspect its output, the prompt that produced it, and where it came from.

The progress view: steps and operations on the left, a grid of documents in
the middle, and the selected document's detail on the right

Turning it on

First install the optional tui extra (it pulls in the Textual library used to draw the dashboard):

pip install "docetl[tui]"

Then add interactive_ui: true at the top level of your config (next to default_model):

default_model: gpt-4.1-nano
interactive_ui: true

pipeline:
  steps:
    - name: themes
      input: reviews
      operations:
        - extract_theme
        - canonicalize_themes
        - summarize_themes
  output:
    type: file
    path: output.json

And run the pipeline the usual way:

docetl run pipeline.yaml

The dashboard only starts when you are in an interactive terminal. In a script, a CI job, or anywhere the output is piped, DocETL automatically falls back to its normal log output, so the flag is safe to leave on.

What you see

There are three panels:

Left — operations. Every step and operation in your pipeline, with a live status, the running counts, the cost, and the elapsed time. The total cost and time for the whole run are at the top.
Middle — documents. One circle per document for the selected operation. A filled circle means the document is done (green), in progress (orange), or errored (red); a hollow circle means it has not started yet. The header shows how far along the operation is.
Right — detail. The output of the document under the cursor, the prompt that produced it, and a short note about where it came from. Documents appear here as soon as they finish, so you can inspect results while the run is still going.

Moving around

Key	Action
`↑` / `↓`	select an operation
`Tab`	switch between the operations list and the document grid
`←` / `→`	move the cursor in the grid (and page through large grids)
`PgUp` / `PgDn`	page through the grid
`Enter`	inspect the document under the cursor
`q`	quit (the run keeps going)

What each kind of operation shows

The progress bar counts the right thing for each operation, and the detail panel adds a short note about where a document came from when that makes sense.

Reduce counts groups, and a group shows how many documents were combined into it:

Split counts chunks, and a chunk shows which piece of its source document it is (for example, "chunk 2 of 5"):

Resolve and equijoin count comparisons as they are made. Their output documents aren't known until the operation finishes, so while it runs the header counts comparisons and the grid shows a ?:

Very large runs

For runs with tens of thousands of documents the grid would never fit on screen, so it switches to a heatmap: each cell stands for a bucket of documents, shaded by how many are done. The live counts and per-operation totals still tell you exactly where things stand.