AI research assistant for analysts

Read faster, see which pages the AI read, and know when a page you rely on changes.

Used by professionals at
  • Accenture
  • Deloitte
Use cases

Three jobs to hand to HARPA first.

01

Answers that show their sources

A finding you can’t trace is a finding you can’t use. Ask the web, and see what the AI read to answer.

  1. Ask in the chat with Web search on (the default with HARPA AI).
  2. Click Searched N websites to see the queries and every page it read.
  3. Pick a Perplexity model when you need a link for each point.
  4. On a PDF, Summarize this PDF, then ask follow-ups.

Good to knowAnswers about a PDF don’t cite page numbers: check a figure in the document before you quote it.

02

Read 40 sources at once

A literature scan asks the same questions of every source. Ask them once and read the answers side by side.

  1. Open your sources in tabs.
  2. Make a command that asks for the claim, the method and the date, as JSON.
  3. Choose Run on tabs…, select Like this page, answer its questions once and Run on N tabs.
  4. Export CSV: one row per page, one column per field.

Good to knowEach answer is capped at 4,000 characters. Bulk runs use HARPA AI cloud credits or your own OpenAI key, not a ChatGPT login.

03

Watch a regulator or journal page

The page your analysis rests on changes quietly. Know the day it does.

  1. Open the regulator or journal page, then the Automate tab.
  2. Set up Page text changes: Changes by more than 5%, or Text appears… with a name you track.
  3. Check every day. Tell me by Email.
  4. Open the monitor to see how much changed, each change in its history.

Good to knowA monitor says that a page changed, not what it means. For that in words, schedule a command that summarizes the page into your Inbox.

Free for desktop Chromium browsers

Put your browser to work.