# Why we built Anchor Terminal

> Vlad Cealicu on why he and Quynh built Anchor Terminal after CryptoCompare, how the benchmark and the agent-first site work, what letme.dev is for, and how the company makes money without selling rankings.

- Canonical: https://www.anchorterminal.com/blog/why-we-built-anchor-terminal
- Markdown: https://www.anchorterminal.com/blog/why-we-built-anchor-terminal.md (~1,350 tokens)
- Slim: https://www.anchorterminal.com/blog/why-we-built-anchor-terminal.min.md (~630 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/blog/why-we-built-anchor-terminal.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (full text: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- Updated: 2026-10-09

## Some context first

I built CryptoCompare in 2014 to make some sense of all the projects in the crypto world and to stay up to date with new coins and technologies, because we were drowning in noise and in new projects being launched. We sold it two years ago, and after my lock-in ended I decided to build something similar for AI agent tools.

Quynh (my co-founder, and our first CPO at my previous company) and I have been looking for project ideas over the past few months. We kept thinking about what we could build in the AI industry that would help. Two weeks ago we came up with [Anchor Terminal](/), a website that we hope makes all the tool and announcement noise easier to understand, and helps you track end of life and tools being decommissioned.

## The numbers, two weeks in

- 961: tools graded, so your agent doesn't have to
- 5,582: head-to-heads, and not one weather API against a crypto feed
- 1,236: reviews signed by agents with opinions
- 31: shutdowns and breaking changes we can already see coming

The directory on 9 October 2026

- 19,339: visitors on our busiest day
- 95,587: requests in a single day
- 71,359: crawler requests on 8 October. Hello, Google
- 1,085: times an agent read the HTML when the Markdown was right there

Our first week live, from our own logs

## The idea

It's hard to stay on top of what is being built in terms of tools for agents, and it's hard to know which ones are good to use. So we devised a [benchmark](/benchmark/), similar to what we did for crypto exchanges ten years ago, to rate tools by how agent-friendly they are and how ready they are to be used by agents.

- 8: reviewer agents, zero coffee breaks
- 7: benchmark categories scored, 2 more waiting on our own probes
- 167: listings good enough for agents (BB or better)
- 386: different things an agent can do, mapped

How the benchmark works today

Each tool is researched by two agents working independently, and an arbiter reviews what they found. A [panel of eight reviewer agents](/reviewers/), each with a different profile, leaves reviews on the graded tools.

## Agent first

We built the website "agent first". In my mind that means every page has a Markdown and a JSON version, and we tried to do anything we can to make the site easy for agents to use. There's an [llms.txt](/llms.txt), a lot of metadata on every page, and everything is open and accessible.

We also built a lot of head-to-head pages, tool X against tool Y. It's been quite successful by my standards. We put the site live a week ago and it reached close to 20,000 unique visitors a day. A lot of that is because we try to create pages for things as soon as they come out. For example, we have pages for Clef, Cloudflare's new open-weight decision model, against all the alternatives. It's the same playbook we used for CryptoCompare, where we always tried to be the first to list anything new in the crypto industry. Our top-performing page so far is [Cloudflare Clef vs Convai Laya](/compare/cloudflare-clef-vs-convai-laya).

## letme.dev

There is another side to the site, a tool for agents only that runs on [letme.dev](/letme/). The idea is that it handles any request an agent has, as a proxy. An agent can call `https://letme.dev/text-to-speech?text=hello&voice=calm` and letme makes the call to the top tool on our benchmark on the agent's behalf.

We'll work on implementing [x402](/x402/) so payments can be automatic, and work with vendors to get a discount when we are used as the proxy. Say we get 5% off, which works as a referral fee. We charge the agent calling us the same as it would pay the provider directly, and letme changes providers when the best one goes out of business, when the agent asks for the cheapest alternative, or when the main one loses points on the benchmark and gets replaced.

We also have a Minisforum S1 Max with 128 GB to test agents with open-weight models, and we're slowly making progress on the proxy and on "choose for me".

## How we make money, and what we won't do

We plan to keep all the APIs open. We make money from audits we sell to companies and from helping companies with their listings, but the ranking can never be bought or influenced.

The vast majority of the revenue we plan comes through the proxy. We can negotiate better deals with the companies that build tools, so your agent pays the same or less, and we get paid by the companies whose tools are used. We plan to be transparent about this, and a company paying us more will never influence its ranking.

We're a team of two with our own servers, so our costs are low. There is no reason to start doing the things traditional companies do and enshittify the product. We're doing this to help the industry grow and stabilise.