Helm.
Helm is Astrolabe's in-house inference layer — a high-performance home GPU tower for hosting large language models, paired with the always-on astrolabe-edge node. Private weights, no rate limits, and real local intelligence for every product we build.
Our own compute. Our own weights.
Large-model class tower
A home 'big tower' built to host and serve large language models at full size.
astrolabe-edge node
An always-on host that keeps inference and agents running around the clock.
Private weights
Models we run on hardware we own — no per-token meter, no data leaving the mesh.
Tailscale-private
Boat, edge node, and tower stitched into one secure private network.
Powers the constellation
PostalListr descriptions, Boat Brain reasoning, and whatever comes next — on our metal.
Hermes-ready
Hermes Agent via OpenRouter today, migrating to local large-model hosting as Helm comes online.
Part of the Astrolabe constellation.
One studio, a set of precise instruments. Helm shares the same engineering DNA as everything else we build.
See the full constellation