Too conservative
Operations take longer than necessary and waste compute, storage and network capacity. A migration sized for the worst hour runs at that rate for all the good ones too.
Adaptive infrastructure orchestration
Infrastructure that adapts to the system it’s running against.
Laminar continuously measures infrastructure capacity, adapts execution in real time, and checkpoints every operation so long-running infrastructure work can survive interruption.
Infrastructure operations should measure the system they are running against, instead of guessing a fixed rate before they start.
prod-db-east · live capacity estimate
ObserveLaminar throughputEstimated safe capacityFixed rate (74 MB/s)
- Throughput
- 66MB/s
- Est. ceiling
- 79MB/s
- Unused at fixed rate
- 11%
- Utilisation
- 84% of est.
Shaded windows are where the fixed rate exceeds what the target can absorb. Laminar reduced its rate instead.
The problem
Infrastructure capacity is not a constant.
Operators traditionally have to choose an execution rate before an operation begins — before anything has been measured. A database under load at 09:00 is not the same target at 03:00, and neither is the network between them. One number, chosen up front, has to cover all of it.
Too aggressive
The target becomes throttled or overloaded, creating operational risk. The work competes with production traffic on the very system it is supposed to be moving.
Laminar addresses this by continuously observing the target and adapting execution to its changing conditions. The rate is not a setting. It is a measurement.
Product
What the control plane gives you.
Five properties that hold for every operation Laminar runs, regardless of which infrastructure it is running against.
Adaptive Execution
Laminar continuously adjusts workload based on observed target capacity, rather than holding a rate that was chosen before the operation began.
Durable Checkpoints
Every operation maintains verified progress, so interrupted work does not have to start from zero.
Automatic Recovery
Operations can recover from worker, network or infrastructure failures, resuming from the last verified checkpoint.
Observable Decisions
Operators can see what Laminar observed, what it estimated, and why it changed execution — each rate change is attributable to a measurement.
Connector Architecture
Infrastructure-specific systems connect through a common connector interface, rather than requiring a separate control system for every infrastructure technology.
How it works
One loop. Seven jobs.
Laminar runs a single control loop against every target it manages. The loop does not stop when execution starts — it is what execution is.
ObserveEstimateProbeLearnAdaptCheckpointRecover
continuous — never a one-shot plan
Observe
stage 1 of 7Read the target's own signals rather than the client's: throttle responses, queue depth, replication lag, latency percentiles, error rates.
pg_stat · consumed capacity units · replication window · p99 latency
Use cases
Seven applications of the same loop.
These are not seven unrelated products. Each one is the same adaptive control loop pointed at a different kind of infrastructure work — which is why a capability added to the loop shows up in all of them.
Migration
Move data between systems at whatever rate the destination can currently absorb.
Validation
Read and compare at the available capacity without starving production traffic.
Automated Repair
Rebuild, re-replicate or backfill against a target that is already under stress.
Right-Sizing
Measure what a workload actually consumes before changing what it is allocated.
Scheduling
Place work in the windows where the target has measured headroom, not assumed headroom.
Region Bootstrap
Stand up a new region by filling the ramp as fast as the new region proves it can take it.
Disaster Recovery
Restore under time pressure, from verified checkpoints, at the maximum safe rate.
Interactive demo
Launch Interactive DemoSee Laminar adapt in real time.
Watch a long-running infrastructure operation respond to changing capacity, adjust its execution rate, and recover from interruption.
- Throughput tracking a live capacity estimate
- Every rate change with the observation behind it
- Verified checkpoints as they are written
- Interruption and recovery, end to end
Technical foundations
Built for environments where the target matters.
Laminar is designed around the assumption that the system being operated on is production, is shared, and will change underneath the operation while it runs.
- Adaptive control
- A closed feedback loop, not a schedule. Execution rate is an output of measurement, never an input chosen in advance.
- Capacity estimation
- An explicit, inspectable estimate of what the target can absorb, carrying a confidence value that decays when signals go stale.
- Feedback-driven execution
- Concurrency and rate move against the current estimate — increasing while headroom holds, backing off on the first sign of target stress.
- Durable checkpoints
- Progress is written as verified markers, so the state of an operation is a recorded fact rather than an inference.
- Resumable operations
- Any operation can be stopped and resumed from its last verified checkpoint, by a different worker if necessary.
- Auditability
- An append-only record of what was observed, what was estimated, what changed, and when — reviewable after the fact.
- Infrastructure connectors
- Each system supplies capacity signals and a rate-control surface through one interface. Adding a system does not mean changing the control loop.