Storage priced for research budgets, not data-center contracts.
All plans include full POSIX access, band-aligned writes, and erasure-coded redundancy. No egress fees. No per-request pricing.
- 10TB usable storage
- 3-node minimum cluster
- Reed-Solomon 4+2 redundancy
- Full POSIX mount
- Community support forum
- SLA guarantee
- Priority support
- 50TB to 1PB usable storage
- 5-node minimum cluster
- Reed-Solomon 4+2 or 8+2
- Full POSIX mount
- Email support, 24h response
- 99.9% uptime SLA
- Monthly usage reports
- Unlimited usable storage
- Dedicated cluster nodes
- Custom erasure coding config
- Full POSIX mount + NFS export
- Dedicated Slack channel
- 99.99% uptime SLA
- On-site deployment assistance
All pricing shown is indicative early-access pricing. Final production pricing will be confirmed before billing begins. No credit card required for Developer tier. Early-access partners receive pricing guarantees in writing before data is committed.
Full feature comparison
| Feature | Developer | Production | Enterprise |
|---|---|---|---|
| Usable storage | Up to 10TB | 50TB to 1PB | 1PB+ |
| Minimum nodes | 3 | 5 | Custom |
| Erasure coding | 4+2 only | 4+2 or 8+2 | Custom |
| POSIX mount | |||
| NFS export | |||
| Uptime SLA | None | 99.9% | 99.99% |
| Support channel | Forum | Email (24h) | Dedicated Slack |
| Egress fees | None | None | None |
| Billing | Free | Monthly | Annual / custom |
Common questions about pricing and access.
Early access means you are using the software while it is still in pre-production testing. We run your workload on hardware we provision and manage, give you a dedicated cluster, and iterate with you on any issues that arise. You are not running a toy demo; this is your real training data. In exchange, we ask for feedback and sign a pricing lock agreement before you commit any data.
Yes. Because Leil presents a POSIX filesystem, migration is an rsync command. We work with early-access partners to plan a migration that keeps the existing storage path live until the Leil cluster is fully populated and read-verified. There is no mandatory cutover window that forces training downtime.
Production tier is billed monthly with no minimum term. Enterprise arrangements typically use annual billing with a discount, but we negotiate this per-deployment. There are no break-up penalties; you export your data via standard POSIX tools and leave.
Standard x86 Linux servers with 8 to 12 SMR drives per node, 10GbE or 25GbE networking, and standard ECC RAM. We specify minimum hardware configurations; early-access partners either procure their own hardware to our spec or use hardware we provision in a Tallinn colocation facility.
Reed-Solomon 4+2 means data is split into 4 data shards and 2 parity shards per stripe. Any 2 of the 6 shards can be lost (failed drives) without data loss. The system continues to serve reads in a degraded mode while rebuilding. 8+2 extends this to any 2 of 10 shards, with higher storage efficiency at scale.
The write scheduler handles mixed patterns. Small random writes are buffered and coalesced; larger sequential writes are passed directly to the band-aligned path. The coalescing buffer introduces a latency penalty on small random writes (typically 50 to 200ms). If your workload is dominated by small random writes at high IOPS, we will tell you upfront that NVMe is a better fit. Checkpoint and dataset workloads are a natural match; interactive databases are not.
We size it together.
Send us your current storage spend, your corpus size, and your checkpoint interval. We will calculate your Leil footprint and send a detailed cost comparison within two business days.