Background tasks

Some things the agent does are instant — a quick answer, a chart swap. Other things take minutes: generating a full PDF report, running an agent loop over a large dataset. For those, Databasin One uses background tasks — you stay in the chat; the work keeps running.

When things go to the background

The agent splits work into two paths automatically.

Fast path

If everything the task needs is already in the current chat — rows already pulled, charts already made — the agent can finish in a single LLM call. That usually takes just a few seconds and stays inline in the conversation.

Examples of fast path:

• "Export this as a PDF" when the chart and data are already in the chat.
• "Summarize what we just found."
• "Make this a one-pager."

Slow path (background task)

When the task needs new analysis from scratch — fresh queries, iterative thinking, a multi-section document — it runs as a background task. That usually takes a minute or a few, and runs independently of the chat.

Examples of slow path:

• "Generate a quarterly review PDF."
• "Write me a 5-section report about pipeline health."
• "Do a full analysis of churn and output it as a document."

The Downloads panel

Background tasks show up in the Downloads panel — a slide-in drawer on the right side of the Databasin One page.

Each task entry shows:

• Status: pending, running, completed, or failed.
• Progress: a rough bar plus the current step (planning, querying, drafting, rendering).
• Actions: download the output, re-run, or delete.

Completed tasks stay in the panel until you dismiss them, so you can re-download a report five minutes after it finished without re-running.

What a background task is actually doing

Under the hood, a background task is just another run of the agent orchestrator, kicked off from a different entry point. It has:

• Its own query budget (defaults to 3, same as chat).
• Its own tool loop — get_schema, execute_sql, skill reads.
• A document-shaped output at the end.

If it fails partway, the panel shows the error and you can retry.

You can keep working

The chat surface stays fully interactive while background tasks run. You can:

• Ask another question.
• Chart a different result.
• Kick off a second background task in parallel.

There's no lock-out. The only thing that shares resources between foreground and background is the LLM rate limit at the org level — heavy parallel usage may slow each task slightly.

When to use background tasks intentionally

You don't have to think about this — the agent picks the right path automatically. But if you want to force a background task (e.g., you know the answer will take a while and you want to walk away), just phrase it as a document:

• "Generate a report about…"
• "Create a PDF summarizing…"
• "Write up a full analysis of…"

Those phrasings route to the slow path.

Try it — open Databasin One→

Where to go next