Full reference

Pair local and remote compute

Pair a compatible running HybrIE machine with your account, keep its heartbeat online, and assign only models its real runtime can serve.

Updated

Reviewed against Console v0.2.2. Available actions depend on project access and runtime configuration; check your installed CLI version for command differences.

Pair a local runtime

Follow Pair local compute for the complete prerequisites, browser approval, model assignment, and troubleshooting. The current entry point is Settings → Organisation → Compute → Pair local compute.

bash
stimulir compute pair \
  --api-base https://api.stimulir.com \
  --runtime-url http://10.0.0.20:8011 \
  --backend auto

Replace the example IP and port with the node base address reachable from the Console backend. The v0.2.2 dialog and CLI accept private IPs or localhost, with no credentials, paths, queries, or fragments. Public tunnel hostnames are rejected. Production cannot reach your laptop through 127.0.0.1; an operator must establish an accessible private network route. A private home-network IP alone is insufficient. If local discovery uses another address, supply --probe-url; keep the registered address reachable from the backend.

Review the CLI pairing code in the browser and approve only your intended machine/project. Keep the CLI and HybrIE running: heartbeats run every 30 seconds. --once registers and exits rather than keeping the machine online. Pairing does not install a runtime or model.

Inspect every remote stage

A workflow can use both local and remote compute when its runtime supports that routing. Check which input, intermediate artifacts and outputs cross the boundary. Verify the selected endpoints before using sensitive data.

Select a model in Models, choose compatible paired compute, and use Assign compute and enable. Wait for Ready, then verify an authorised request and its runtime evidence. Clef/Flash on Metal requires a verified MLX runtime with decision support and installed weights; NPU pairing alone does not run the current Clef workers.

Confirm supported peer behavior

Peer discovery, artifact distribution and remote execution are different capabilities. Use the runtime’s current capability and health information to establish which are enabled. Do not infer remote execution support from a peer appearing in a list.

Private pairing does not share the machine with other businesses. Shared peer compute requires separate owner consent, administrator pool membership and applicable access grants, plus runtime reachability and admission. Sharing is not monetised; cloud calls and rented GPUs are separate charges.

Keep identities and service endpoints distinct

A device identity registry is not the same as the compute service peer registry. Manage executable runtime endpoints through Workers. Check Deployment choices for the distinction between hosted access and an operator-managed runtime.

Command catalogue

Expand a command for its arguments. These commands follow the released CLI source; check stimulir --version and the command’s --help for your installation. Required flags are marked; some operations also validate combinations of inputs.

stimulir compute pair

Pair local CUDA, Metal or NPU compute with a project using browser approval.

--runtime-url required
HybrIE node address reachable from Console, without /v1.
--api-base
Console API address.
--once
Register once without heartbeat polling.
--probe-url
Local runtime address for capability discovery; registration still uses --runtime-url.
--backend
auto, cuda, metal or npu.
--memory-gb
Declared memory for Metal/NPU when automatic detection is unavailable.