Back to the corkboard

Ansa

◐ Sprout Planted #agents #projects

ansa.alexytsu.com is a research tool where every question builds a high-quality article written by a researcher agent, pinned to sources and interrogated by a series of questions. Synthesis is driven by posing questions to the agent via a “comment” section.

Ansa showing an article on Borges' The Library of Babel, with numbered footnotes, a highlighted passage, and an agent's answer in the comment column Ansa showing an article on Borges' The Library of Babel, with numbered footnotes, a highlighted passage, and an agent's answer in the comment column
An article with footnotes. The highlighted passage has a comment thread alongside it.

Why this exists

When I’m digging into a research topic I always end up wanting to drill into specific parts of a response. Forking the conversation is sort of what I want, but each fork becoming its own top-level chat is poor UX.

I think that the ideal experience for these branching conversations is comment trees (think Reddit or HackerNews).

  • Each thread gets clean context. By default, the agent sees the article and the chain of comments it’s replying to, but nothing from sibling threads, so one tangent never muddies another.
  • It’s much lighter than a conversation fork. Threads can be collapsed or hidden, and you can skim several forks in one view instead of jumping between whole “chats”.

I tend to buy into the narrative that “AI is making us dumber”. I’m happy to let AI write my code, but I don’t want it doing my understanding for me. Knowledge you can recall and reason with has to live in your own head. Ansa is designed to help understand things:

  • The back and forth process is more “active” than reading a wall of LLM output
  • You can drill down on specific things you don’t understand in an efficient manner

The articles are worth keeping for yourself, but I wouldn’t suggest sharing them widely (but do what you want). Most of the value is in the process of generating and interrogating them, imo, and an article without that process behind it is just another wall of LLM output. I’d rather see this used for “personal knowledge graphs” than for sharing knowledge within organizations.

What to use it for

  • Create high quality research articles driven by the dialectic method and with a fully auditable trail
  • Learn technical things. Keep asking questions and diving deeper on areas of interest or gaps in understanding.
    • Great for diving into sprawling topics where you are unfamiliar with some domains
  • Learn a new language
    • I get it to generate easy reading articles in Japanese tailored to my level and then I can ask for translations, clarifications or explanations of what I don’t understand
  • As a general entry point for any time you want to ask ChatGPT or Claude about something you’ll want to refer back to later
    • Planning a holiday
    • Documenting your life admin
  • Anything you want… (I’m not your boss)

How to use it

Start with a prompt that generates the initial article/post. This provides some anchoring and context for the agent to reference in all threads scoped to that session.

Highlight any text in the article or a comment and write more prompts. You have three modes to choose from.

1. Ask

The agent replies with a short answer, right in the thread where you asked.

An Ask comment on the article's lead, “Who first translated it into English, and when?”, answered by an agent with the 1962 translations by Irby and Kerrigan and a link to the Publication history section An Ask comment on the article's lead, “Who first translated it into English, and when?”, answered by an agent with the 1962 translations by Irby and Kerrigan and a link to the Publication history section

2. Deep dive

The agent spins off a new article on that tangent, linked both ways.

A Deep dive comment asking to go deeper on Greek mythology, with the agent's reply and a card linking to the new Greek mythology article it spun off A Deep dive comment asking to go deeper on Greek mythology, with the agent's reply and a card linking to the new Greek mythology article it spun off

3. Edit

The agent proposes a change to the article, which you review and then accept or reject.

An Edit comment asking to tighten the article's lead, with the agent's reply marked edit accepted An Edit comment asking to tighten the article's lead, with the agent's reply marked edit accepted

Mobile UI

For on-the-go use

The end of the Library of Babel article on a phone, with its spinoff article card, followed by the comment section and the first comment thread The end of the Library of Babel article on a phone, with its spinoff article card, followed by the comment section and the first comment thread
On a phone, the comments sit beneath the article.

Try it

Sign-in is invite-only. Send me your email and I’ll give you access.