Husk Mesh Vault provisions private bare-metal and software-defined network infrastructure for engineering teams in Japan who need full control over their compute, storage, and routing layers.
Dedicated bare-metal nodes running on Intel Xeon (4th gen) or AMD EPYC 9004 series processors. No hypervisor is installed by default. You receive IPMI access, a clean OS image of your choice (Ubuntu 22.04 LTS, Debian 12, or Rocky Linux 9), and a 10 Gbps uplink to the facility's core switch.
Software-defined layer-2 and layer-3 networking configured via a versioned REST API. Supports VLAN segmentation, static routes, BGP peering with your own ASN, and per-port ACLs. Every change is recorded in the audit log with a timestamp, the authenticated user, and the previous and new values.
S3-compatible object storage built on dedicated NVMe nodes with erasure coding (8+4 by default). Supports bucket versioning, lifecycle policies, and server-side encryption with AES-256. The API endpoint is accessible only over your private VLAN; it is not exposed to the public internet.
An immutable append-only log of every event on the platform: API calls, SSH logins, configuration diffs, and billing events. Logs are signed with a platform key and can be exported in JSON or CEF format to your SIEM. Retention is 13 months by default, configurable up to 7 years for regulated industries.
Kenji Arakawa spent eight years building distributed systems at a Tokyo-based financial data provider, where the infrastructure ran on a major US hyperscaler. The arrangement worked until it did not. In late 2018, a billing anomaly triggered a support ticket that took eleven days to resolve, during which the team had no visibility into what was actually running on their behalf. That incident was the direct reason Husk Mesh Vault was incorporated in January 2019. The premise was simple: engineering teams should be able to see every layer of the stack they depend on.
The first year was mostly hardware. Kenji negotiated a colocation agreement with a facility in Chiba, racked the first twelve servers himself, and wrote the initial provisioning API over a long summer. The first paying customer was a three-person fintech team from Osaka who needed bare-metal Kubernetes nodes that their compliance officer could physically audit. That constraint, the ability to hand an auditor a floor plan and a cable diagram, shaped every subsequent design decision. By the end of 2020, the platform had grown to 48 nodes and four customers.
The choice between a bare-metal server and a virtual private server is not primarily a performance question. For most engineering teams in Japan, it comes down to two things: what your compliance requirements say about shared tenancy, and how much of the stack you need to be able to explain to an auditor.
Read more →Network configuration has historically been managed through CLI sessions on individual devices, with changes…
S3-compatible object storage has become the default interface for storing unstructured data. The API is…
Spin up dedicated nodes in our Chiba facility within 90 minutes. No hypervisor overhead, no noisy neighbours, no shared tenancy.
Configure VLANs, BGP peers, and firewall rules through a versioned API. Every change is logged, diffable, and rollback-capable.
S3-compatible vault storage with AES-256 encryption at rest and TLS 1.3 in transit. Keys stay in your control plane, not ours.
"Kenji Arakawa studied computer engineering at Tohoku University and spent eight years as a senior infrastructure engineer at a Tokyo financial data firm, where he was responsible for the distributed storage layer serving real-time market feeds. He left in late 2018 to found Husk Mesh Vault after a prolonged cloud-provider support incident made clear that opacity was a structural feature, not a bug, of shared infrastructure. He holds a CISSP certification and reads IETF RFCs for what he describes as light reading on the train to Chiba."