Didier FRANC

Posted on May 20 • Edited on Jun 3

Agnostic, A Startup Story 📚

#startup #ai #blockchain #data

Agnostic: The Story of Two Engineers Who Wanted to Make the Blockchain Understandable

At first, it was just two friends obsessed with data. Today, it’s an open-source stack used to explore blockchains like databases. This is the story of how we got here.

Chapter 1 — Hitting the Wall

It all started the way these things often do: with frustration.

It was the end of summer 2021. Arnaud and I had just started working together on what would become Agnostic. We were building tools to explore Ethereum, Polygon, and Arbitrum—not dashboards, but low-level data exploration systems. We wanted to answer basic questions:

Who interacted with this contract? What’s the full path of this transaction? Where did the ETH go?

But again and again, we hit the same wall.

Blockchain data was opaque. Fragmented. Slow. We stitched together RPCs, decoded logs manually, exported to spreadsheets, ran fragile ETL pipelines.

We kept asking ourselves:

Why is it so hard to query blockchain data like regular data?

That’s when the idea for Agnostic was born.

We didn’t want to build just another dashboard or indexer.

We wanted to build a language—a way to talk to the chain.

Chapter 2 — Bootstrapping the Hard Way

We had no funding. Just time, passion, and stubborn conviction.

We hosted our own nodes, called them billions of times. We optimized everything to run on minimal infra. Every request was budgeted. Every query was tuned. Every delay became a lesson.

We ran ClickHouse on bare metal. We scraped chains in parallel. We decoded hex manually before building our first parsers.

It was duct-tape engineering—but it worked.

It was fast. It was raw. And it was ours.

Chapter 3 — From AI to Blockchain

Before Agnostic, Arnaud and I had worked together on AI projects—tools for generative media and intelligent data pipelines. That’s where our trust as co-founders was forged.

Yann, who had been my intern back in 2016, joined us later. He and I had already built tools for Deloitte, a bidding platform for rail, maritime, and aviation, and a generative video editor before blockchain.

When he saw how far Agnostic had come, he didn’t ask for a pitch.

He just asked:

“When do I start?”

That kind of loyalty is rare—and it became part of the DNA of the team.

Chapter 4 — Reinventing Decoding

Our first decoding engine was huge—a PostgreSQL database packed with every function and event signature we could find. It powered a high-speed chain processor that worked well… but wasn’t scalable.

Then we had a breakthrough:

What if everything needed to reconstruct an ABI could fit into a single string?

That’s how we created the full signature—a compact format that turned gigabytes of decoding metadata into a few megabytes.

This unlocked:

On-the-fly decoding in ClickHouse using UDFs
Stateless, embeddable pipelines
SQL-only ETLs and lightweight orchestration via AGT (Agnostic Transform)

Chapter 5 — From MergeTree to Iceberg

Our original backend used ClickHouse MergeTree tables—blazing fast, but not designed for open data or multi-engine access.

So we made a big bet: we migrated everything to Apache Iceberg.

It wasn’t well supported by ClickHouse at the time, so we built it ourselves:

Custom compaction & indexing
Partitioning strategies for multi-chain scale
Seamless integration with Spark, DuckDB, Trino

The result: we turned our internal datasets—logs, traces, transactions, and blocks—into public, versioned Iceberg + Parquet datasets hosted on S3.

Free. Composable. Ready to use by anyone.

Chapter 6 — Breaking the Monolith

At first, everything was bundled together—explorer, decoder, ETL, infra.

But the more we listened to users, the more we realized they needed modularity. So we broke it all apart.

The explorer became AGX
The live decoding engine became clickhouse-evm
The pipelines became declarative SQL, orchestrated by AGT
The data became public and pluggable

Agnostic became a stack, not a product.

Chapter 7 — AGX Becomes Its Own Thing

AGX started as a simple UI to browse decoded on-chain data.

But it grew fast. We realized we were building something more powerful: a universal data exploration tool.

We rebuilt it as a native app, using Tauri + ClickHouse local + Svelte.

Fully local, fully offline
Save queries, charts, dashboards
Run analytics on blockchain or CSVs or your own data warehouse

It’s fast. It’s yours. And it’s not just for blockchain anymore.

Chapter 8 — AI for Everyone

Even with great tools, we hit one last barrier: SQL.

Not everyone can write it. Even engineers struggle. So we added AI.

We trained, tuned, and embedded text-to-SQL into AGX. Today, it can:

Convert natural language into optimized SQL
Fix broken queries
Autocomplete based on schema and data context
Help anyone ask better questions—even without technical skills

This was the unlock that made Agnostic accessible to everyone.

Chapter 9 — John Calabrese Joins the Mission (2024)

By early 2024, Agnostic was powerful. But we wanted to test its limits. We needed someone who would push the product beyond what we imagined.

That person was John Calabrese.

After an impressive run at Nansen, John joined us—not just as a user, but as a builder. He broke the product. He bent it. He used it in ways we hadn’t designed for. And in doing so, he showed us its true potential.

He helped turn Agnostic from a product into a platform. And he was the architect of our notebook platform to share your insights with the community

Chapter 10 — From Passion to Funding

We worked unpaid for over a year. No salaries. No grants. Just fire.

But the ecosystem noticed. Devs shared dashboards. Protocols reached out. Data teams integrated us.

Then came our first offer—from BlueYard. They believed in the long-term vision and led a €2M round.

We were joined by:

Kima Ventures
Stake Capital
TRGC
Atka

We didn’t celebrate with champagne—we shipped another feature.

Epilogue — This Is Just the Beginning

We didn’t start with a plan. We started with a feeling:

This should exist.

Today, Agnostic is:

A modular open-source stack
A public blockchain data lake
A blazing-fast local explorer
A SQL-native, AI-augmented toolchain

Tomorrow, it could become the standard for how we explore on-chain data.

We’re not trying to build everything. We’re building the parts that let everyone else build.

DEV Community