From The Graph's First Hire to Verified Onchain Data: Rodrigo Coelho on Amp

Around 2016, Rodrigo Coelho was working out of a co-working space in San Francisco when he started talking with another founder nearby. "He had like an Ethereum sweatshirt on, so we started talking," Coelho recalled. The founder was Yaniv Tal, and the pitch deck he was working on was for The Graph. Coelho had been starting tech companies since the dot-com era, so he offered to introduce Tal to investors. A few months later the project got its first check. Tal asked him to join, and Coelho became the first hire in 2018.

Coelho is now CEO of Edge & Node, the company that grew out of that original team. Speaking with DevNTell host Narb, he walked through Amp, Edge & Node's newer product. Amp turns blockchain activity into structured, verifiable datasets that can be queried with SQL.

The First Hire at The Graph

Coelho described his early role as "an adjacent ops person" working alongside three co-founders. He worked on community and helped plan Graph Day. That conference happened to be Narb's first crypto event, and Narb won prize money at its hackathon at the Palace of Fine Arts in San Francisco. Coelho mentioned that an even smaller first Graph Day had come before it, held in a theater.

The Graph protocol took about three years to build. It launched in 2021 as a decentralized network that indexes blockchain data and lets developers query it. The company began as Graph Protocol, Inc. and renamed itself Edge & Node when the network launched. According to Coelho, the infrastructure has served trillions of queries over the years, still serves billions a month, and runs behind many DeFi protocols. Its token, GRT, is still in use.

The early years were not smooth. Coelho pointed to unclear SEC regulation at the time, when "the state of the art back then was to set up like, you know, a Cayman Islands foundation" and launch a token from there. He also named the work of keeping infrastructure running after usage took off immediately, and the work of growing the staff.

Banks Running the Open Source Code

Amp started with something Edge & Node heard secondhand. In 2021 and 2022, financial institutions began experimenting with blockchain data. The Graph's code was open source under the MIT license, and word came "through the grapevine" that large institutions had downloaded the software and were running it internally.

With more research, Edge & Node concluded that subgraphs had been built for one particular kind of developer. Institutions needed something else. They already had data analytics teams and pipelines, and they wanted blockchain data in Snowflake, BigQuery, Oracle, or Postgres. Edge & Node spent a couple of years in R&D on extraction that keeps up with each chain and handles the quirks of blockchain data. It then wrapped that in a tool where data engineers write SQL to filter the data into datasets and send those datasets where they need to go.

Choosing SQL over GraphQL, which subgraphs had used, was deliberate. Coelho called SQL "kind of the most broadly known language in the world." He gave a second reason too: "AIs natively speak SQL, so that's another benefit."

A Preview of Amp Cloud

For the demo, Coelho showed Amp Cloud, a visual interface that has not been released yet. Data engineers usually run Amp itself from the command line or from Python notebooks. Amp Cloud is meant for people who are not data engineers. He typed "Show me like USDC transfers on Base and Ethereum." An LLM picked the tables, wrote the SQL query, and showed a preview of ERC-20 transfers on both chains.

The query could be copied or edited, then saved as a dataset. Coelho described a dataset as a database table that stays current: each new block adds a row if it matches the criteria. A dataset can run once or keep syncing block by block.

He then turned to reorgs, where a chain backs up and continues down a different fork. When that happens, data that was already exported no longer matches the chain.

So if you're dealing, for example, with like financial institutions that are doing like needing accurate data for like to make trading decisions, right? You can't have data be inaccurate and not match what's onchain.

Amp handles those rollbacks itself. Exporting a dataset creates a pipeline, and the interface draws it from left to right. The sources sit on the left (two chains in the demo), a transformation sits in the middle, and a destination such as a BigQuery warehouse sits on the right. The interface also keeps a user's own datasets apart from derived datasets that other people in the organization have made, so nobody has to rebuild a table that already exists.

The Verified Stamp

Coelho paid particular attention to a small "verified" label in the interface. Amp produces a cryptographic proof that stored data matches what is onchain, checking it against block headers and the chain's Merkle roots. The label is aimed at audit and compliance teams that want to spot check a data warehouse or pull a report.

Coelho framed it as a garbage in, garbage out problem. Nodes can drop data, so a dataset might be missing a block somewhere. If a large transfer was in that block, any total that includes it will be wrong. "For financial information, it's got to be 100% accurate and complete," he said. Edge & Node's product documentation draws a line here: the checks catch inconsistent RPC responses, but they do not establish consensus finality or prove that downstream business logic is correct.

The way Amp is deployed is part of the same pitch to institutions. Amp, the engine, is sold as licensed software that runs inside the customer's own secure cloud, which Coelho compared to licensing Oracle. He described the typical buyer as a large company where paying for data by consumption has become expensive and the "build versus buy" question has come up. Amp Cloud is planned as a monthly subscription with a free tier. It is meant as a first step for smaller users, and for institutions running proofs of concept before they bring Amp in-house.

Micropayments Before x402

Narb asked whether agents might pay per request for Amp data through x402. Coelho said Edge & Node was considering pricing models for Amp Cloud, and he noted that The Graph already has a Graph MCP. Edge & Node's interest in micropayments goes back to 2021. The Graph handled billions of queries that each cost less than a cent, and paying gas on each one made no sense because "the gas itself was more than the query itself." Edge & Node built a batching system that is now used within The Graph.

In late 2024, Edge & Node started researching how agents might pay. When x402 came out, the company adopted it early, contributed alongside the Coinbase team, and met with Google about its A2A work. That work became ampersend, a platform for monitoring and controlling payments in agentic commerce. It includes a rules and policy engine and runs on Amp behind the scenes.

Coelho also mentioned a recent talk at a data management summit in New York. His argument was that AI produces large amounts of new data, which raises storage costs, while also making custom-built infrastructure cheaper to develop. Edge & Node applied that idea to its own systems. "We cut like our cloud bills by like 84% over time," he said, and he described Amp as the company's first example of purpose-built data infrastructure.

Amp Cloud has no self-serve signup yet. Until it launches, Coelho said, access goes through a "Talk to us" button on edgeandnode.com and a scheduled call. Self-serve registration and a free tier are planned for later.