An internal research desk with a searchable answer archive

I want an internal research desk that anyone at the company can open instead of pinging the team with "can someone look this up". Build it as an app with an ask screen, a live queue, an answer view, and a searchable archive.

The main screen is the queue: every submitted research question, newest first, showing the question, who asked it, whether they chose quick or deep, the current status (checking, running, answered, failed), when it was submitted, and what it cost. Clicking a row opens the full answer. Keep the queue and its in-flight state in the app's own stored data so a refresh or a second viewer sees the same list, and file completed answers into Notion as described below.

Submitting starts on an Ask screen. The person types their question, enters or picks their name, and chooses Quick or Deep before they submit. As soon as a question has been typed, and before the submit button does anything, the app searches the existing archive using Notion Query a Data Source against the answers database and shows any close matches inline under the box: the earlier question, the date it was answered, who asked it, and the opening of the stored answer. This dedupe step is the whole point of the app, so make it prominent. If there is a match, "Use this answer" is the primary action and starting a new run is the secondary one. Only spend credits when the person explicitly chooses to.

Quick questions run through Exa Answer, which returns a written answer with citations in a single call. Deep questions go to Exa Create Agent Run with a higher effort tier and a budget cap on maximum cost in dollars, and the app polls Exa Get Agent Run while the queue is open until the run reports completed or failed, updating the row's status as it goes. Use a low effort tier for Quick and a high one for Deep, expose the exact tiers and the spending cap in a settings screen, and treat a failed or budget-exhausted run as a visible failed row with the reason on it, never a silent drop.

For every source behind an answer, fetch readable text with Exa Get Contents so a reader can inspect a citation without leaving the app. Render answers so each claim sits next to the sources that support it: inline numbered markers in the answer text, a sources panel beside it listing title, domain and link, and an expandable panel holding the extracted text for each source. Never render an answer as a wall of text with the links dumped at the bottom.

When an answer completes, file it into a Notion database with Notion Create a Page: the question as the page title, and properties for the requester, the date, the effort tier used, the cost in dollars, and the source URLs, with the full answer body and the source list written as page content. Let the person add a short topic tag before it files. If the target database is missing a property the app needs, say which ones to add rather than failing silently.

The archive screen reads back from Notion with Query a Data Source, with a search box over the questions and filters for requester, date range and tag. Opening an archived entry shows the stored answer and its sources the same way the live answer view does. Any colleague can correct or annotate a stored answer from that screen: pull the page's blocks with Retrieve Block Children, let them edit the answer text or add a correction note, and save it back with Notion Update a Block, recording who changed it and when.

Cost and effort belong in the open, not buried in a detail view. Show the cost figure Exa returns on each answer, agent run and contents fetch, show the effort tier beside every question in the queue and on the answer page, and put a running total for the current week at the top of the queue so the team can see what the desk is spending.