Private job runner for teams

Run any job.
On your own infra.

labhub.run schedules, isolates and observes your workloads (builds, ML training, data pipelines, security scans) on the machines you already own. One control plane, zero data leaving your perimeter.

~/acme/ml

$ labhub run train.yml --pool gpu-a100
→ resolving pool gpu-a100 (eu-paris-1, 8 runners online)
→ job j_8f2c41 scheduled on runner-04
→ pulling image registry.acme.internal/train:2.4 … cached
  epoch 1/3  loss 0.412  ████████░░░░  
  epoch 3/3  loss 0.087  ████████████  
✓ succeeded in 14m02s · artifacts → s3://acme-models/j_8f2c41
$

Product

Your hardware. Our orchestration.

Install a lightweight agent on any Linux host. labhub.run turns your fleet into a single, queryable pool of compute.

Bring your own infra

Bare metal, on-prem racks, any cloud or air-gapped networks. The agent dials out, so no inbound ports are needed.

Smart scheduling

Queues, priorities, quotas and labels. Jobs land on the right runner: GPU, ARM, high-memory, or that one licensed box.

Isolated by default

Every job runs in a fresh, sandboxed container with scoped secrets and network policies, torn down when it's done.

Live observability

Streaming logs, metrics and artifacts for every run. Full history, searchable, exportable to your own stack.

API & CLI first

Trigger jobs from CI, cron, webhooks or your own code. Everything in the dashboard is available through the API.

Teams & billing

Projects, SSO, RBAC and per-team usage reports, so finance knows exactly which team burned which GPU hours.

How it works

From zero to first run in an afternoon.

Install the agentOne static binary per host. It registers to your org and reports its capabilities.
Describe the jobA short YAML file: image, command, resources, secrets. Versioned next to your code.
Run it anywhereFrom the CLI, the API or a schedule. labhub.run picks the runner and streams results back.

train.yml

name: nightly-training
pool: gpu-a100
image: registry.acme.internal/train:2.4
run: python train.py --epochs 3
resources:
  gpu: 1
  memory: 64Gi
secrets: [HF_TOKEN, S3_KEY]
schedule: "0 2 * * *"
artifacts: s3://acme-models/

Security

Built for teams who can't ship data out.

Data stays home. Code, data and artifacts never touch our servers. Only job metadata does.
Outbound-only agent. mTLS to the control plane, no open ports on your side.
Self-hosted control plane. Available for regulated and air-gapped environments.
Audit everything. Who ran what, where, with which secrets, immutable and exportable.
Written in Rust. Memory-safe agent, small attack surface, reproducible builds.
EU-hosted. Managed control plane operated in the European Union.

Run any job.On your own infra.