Hot storage performance, cold storage pricing.

Write 1 TB$1
Read 1 TB$10
Store 1 TB for a month$2.50

The base unit of the network, and therefore of Tessera, is the slab: 40 MiB of logical data, erasure-coded into 30 sectors of 4 MiB each and distributed across 30 independent hosts. Any 10 of those sectors rebuild the slab. You can hold terabytes and still read back a single slab — paying the read rate only for the slabs you actually retrieve. A slab costs about $0.000042 to write, $0.000105 to keep for a month, and $0.000419 to read back.

Typical "critical backup" jobs, priced.

Rates below use the current card (storage $2.50/TB-month, uploads $1/TB, downloads $10/TB). A 40 MiB slab is the billing quantum; the costs shown already include it.

What you protectLogical sizeMonthly storageOne write + selective read
One skills pack (100 files, ~10 KB each)~1 MB$0.000002$0.00001
100 skills files, each stored as its own object (full slab quantum)~1 MB~$0.01~$0.05
Skills + project docs + a repo snapshot~50 MB$0.00012$0.0006
1 TB research corpus, 10 slabs retrieved1 TB$2.50$1 write + $0.004 read
50 TB generated dataset archive, 1 TB of selected slabs retrieved50 TB$125$50 write + $10 read

Even if every item above is stored as its own object and the slab quantum is paid in full, the monthly total for a complete agent working set remains well under a dollar. Packing small files (folder upload or tessera sync) removes the per-object quantum and brings the numbers to the logical-volume column.

The comparison that matters

The cost of losing any of these items — a skills library, a decision log, a set of credentials, a repo that cannot be cheaply recomputed — is not measured in fractions of a cent. It is measured in recovery time, re-work, or permanent loss. At these rates the protection is cheaper than the downside of not having it.

Volume and bulk uploads

A bulk upload is limited by your client's bandwidth and CPU (encryption and erasure coding happen locally). On a 1 Gbps link, 100 TB is on the order of days, not hours. The CLI does not publish a guaranteed throughput; measure with a representative file before committing a large archive.