Ardent keeps database branching for coding agents fast and dependable with Raincurve
Ardent (Y Combinator P26) gives coding agents isolated copies of production Postgres in under six seconds. Raincurve brings cross-layer reliability intelligence to the infrastructure behind every clone.
| At a glance | |
|---|---|
| Organization | Ardent (Y Combinator P26) — database branching for coding agents |
| Use case | Cross-layer reliability intelligence for a fleet of on-demand Postgres clones |
| Raincurve capabilities | Cross-Layer Infrastructure, Causal Incident Intelligence, Failure & Impact Prediction |
| Related solution | Enterprise Infrastructure |
The challenge
Ardent lets coding agents test database code against a 1:1 copy of production Postgres. Its promise is specific: clones created in under six seconds, isolated from production at both the compute and storage level, scaling to zero when idle and never duplicating storage, even at terabyte scale.
Each of those promises depends on several layers behaving together: the source database and its provider (Supabase, AWS RDS or PlanetScale), the storage layer shared across clones, and the compute that scales up for agent workloads and back down again. When a clone is slow or fails, the symptom shows up in one layer while the cause often sits in another.
Agents also create clones continuously and at machine speed. The volume of events grows with every customer's agents, faster than any team can investigate by hand.
The approach
Raincurve runs as a reasoning layer above Ardent's existing monitoring:
- One dependency graph linking every clone to its source database, provider, storage and compute, kept current as clones are created and torn down.
- Causal incident grouping so a slowdown in clone creation is traced across storage, compute and provider signals to a single incident with a ranked origin.
- Failure and impact prediction that flags trends threatening clone speed or isolation before customers' agents feel them.
- Verified changes so infrastructure updates are checked against live clone load before they roll out.
What we are measuring
The engagement is tracked against outcomes agreed up front:
- Clone creation time against the six-second target
- Mean time to resolution for clone failures
- Share of incidents with a correctly identified origin
- Changes blocked or re-sequenced by verification before they caused harm
Results will be published here as the deployment matures.
Why it matters
Coding agents create infrastructure demand that no human team schedules. The reliability layer underneath them has to reason as fast as the agents act, which is the problem Curve-1 was built for.
More customer stories
COSMED runs Raincurve in production for continuous log monitoring and incident intelligence
Data centers, carriers, AI infrastructure and enterprise networks.
Private, hybrid or cloud — always inside your boundary.
Start building on Raincurve.
Start with a scoped assessment of your topology, telemetry and recent incidents.