2090 Labs / Founder
Abhishek DG
I build Chronos, the database where every AI agent works in its own copy of your data, and changes land only by your rules or your approval. Before it, TailsMart and ACRED India.
- Role
- Co-founder & CEO, 2090 Labs
- Based
- Bangalore, India
- Building
- Chronos DB, Chronos Gateway
- Before
- TailsMart, ACRED India
Path
Two companies built. Both stopped for this one.
I'm not a database veteran. I built products for real businesses, and the data layer kept becoming the problem.
- beforeTailsMartstopped
Founder and CEO.
- beforeACRED Indiastopped
Where the problem showed up: the moment an AI agent can write to real data, one bad run can wreck it.
- beforeAI-native commercestopped
ConvoSuite and an AI-native Shopify alternative. Too much to finish alone, and it taught me to pick one narrow piece.
- First commit of Chronosday 1
A first released, tested engine nine days later.
- alsoCairnon npm
A shared, append-only journal, so agents working in parallel don't duplicate or undo each other.
- Chronos DB 0.4.8, Gateway 0.2.9latest
Signed releases you install and run yourself.
How I work
The tests are the spec. The rest is typing.
Agents write most of the code.
Most of Chronos is written with Claude Code, several agents at once on separate branches. I set the design and the tests a change has to pass.
Tests decide what lands.
Differential tests against real Postgres, SQLite's sqllogictest corpus, crash tests that cut the log at random bytes, and kill -9 during concurrent merges.
Say where it falls short.
Benchmarks include the ones Chronos loses. If it isn't tested on your setup yet, the page says so, because that is what you would want to know first.
What's the write you'd least want an agent to get wrong?
Tell me. I'd like to test against it.