Today marks a special day for us!
We’re thrilled to announce our $2.1M pre-seed round to build the next layer of data collaboration between humans and AI agents
Huge thanks to our backers @SchematicVC, @skydeck_berkeley, and angels from @datadoghq , @OpenAI , @MistralAI & more
@Polymarket Someone trusted AI a bit too much. Working with AI Agents with no guardrails is like doing a trust fall from the empire state building, the safety nets exists tho and some are even open-source: https://t.co/G9GlNvKASR
We might have accidentally made the first digital hair loss prevention product for developers. And it's open-source.
The classic 3-in-1 is slammed on it: Instant DB branching, Instant rollback & versioning.
Working with AI agents is unavoidable, so at least do it stress-free.
For all the release managers out there, listen:
Fresh DB clone for every PR, auto-provisioned.
Validate migrations against production-identical data.
PII masked at clone time with compliance by default.
Music for your ears? Then check https://t.co/aNCzfykLHS 👈
Alexey Grigorev (https://t.co/1MVlqs0RoY) gave Claude Code access to Terraform. The agent thought the infra looked wrong… and wiped everything .
In this demo, Yassine Ghorbel shows how GFS prevents that agents run on isolated database branches, not production.
AI agents are smart, yet most teams still keep them away from their database.
That's where GFS comes in.
Every agent gets its own branch, its own sandbox to experiment in.
It can safely try, break, and fix things without any risk of data loss.
This thumbnail is from a parallel universe where MrBeast is a tech bro.
What’s not from a parallel universe is Claude Code deleting a 2 and a half years database.
Ask Alexey, the founder of Datatalks,he felt that. Avoid this bad spot, work smart, use GFS https://t.co/P7Vn3zcsq9
3 reasons write agents scare prod 🧵
1/ Silent corruption > loud failure
Bad writes sit quietly wrong for days.
2/ Agents know schema, not invariants.
One delete lead to chaos later.
3/ Rollback is an illusion.
Other systems already read the data
Nah.
AI agents can't be trusted with a shared staging DB that 3 other teams already used this Morning 🙂
Fix the environment before you blame the agent.
Isolated Realistic. Safe. https://t.co/DRTi9Z27Gb 😏
#AI#DataEngineering
production is never even in the conversation 😌
Write whatever, break whatever. No risks no consequences.
Your Agent is done? The branch disappears, Like it was never there.
This is the Guepard experience
You didn't ship a limited agent because of the model.
You shipped a limited agent because you were too afraid to test on real db The database was always the ceiling. 😏
We're building the fix at https://t.co/DRTi9Z27Gb curious what your current work around looks like ?
Truth bomb: AI agents are 100% useless… until you connect them to real tools/APIs. No actions = Just expensive autocomplete.
Give them hands (function calling, Zapier-style integrations) and watch them actually work.
Who's building agents with proper tooling? 🐆
#AI#AIAgents