Spin Up Cells

Choose hosted or cluster act, connect compute, define cell profiles, and run workflows with policy at birth.

Open ComputeCell profiles

Choose hosted or cluster act

A run keeps the same workflowId you registered earlier. Hosted Fast act omits clusterId. Production isolated act targets a connected compute target and cell profile.

Where act runs

Pick hosted Fast act or connect a cluster for isolated act. Blekline does not operate your control plane or worker fleets.

Hyperscaler Kubernetes

  • Amazon EKSdeploy/terraform/blekline-eks installs sidecar, optional admission chart, and compute agent; then cluster heartbeat.
  • Google GKEdeploy/terraform/blekline-gke; same pattern as EKS (sidecar, optional admission, heartbeat).
  • Azure AKSdeploy/terraform/blekline-aks wires the Blekline sidecar into your existing AKS cluster; extend with admission like EKS when you enable it.

Platform engineering

  • Existing clustersBring any CNCF cluster; install Blekline sidecar (and admission on EKS when install_admission is true).
  • Cell pool operatorIn-cluster dispatcher uses operator token for register, claim, and run lifecycle.
  • Policy bundleGET /api/compute/policy-bundle keeps act cells aligned with published runtime policy.

Hosted act (no cluster)

  • Fast actTeam tier hosted runtime uses system profiles such as blekline-small; omit clusterId on cell-runs.
  • Dev sandboxesPartner dev VMs (Daytona, E2B, Modal) are for exploration, not governed Blekline act. See Dev sandboxes vs production cells.
  • Same workflow idRegister workflows the same way; only placement and profiles change.

Partner dev sandboxes are not cell profiles. Dev sandboxes vs production cells explains when to use each.

Register a compute target

In Compute, create a target and copy cluster_id, workspace id, and registration token. Your in-cluster agent calls POST /api/compute/cluster-heartbeat after Helm or Terraform install.

cluster-heartbeat.json
{
  "workspace_id": "ws_your_workspace_id",
  "cluster_id": "ct_your_compute_target_id",
  "registration_token": "your_registration_token_from_console",
  "agent_version": "1.0.0"
}

Install in the cluster

Use the thin Terraform wrappers for your cloud (existing cluster required). Blekline does not create your Kubernetes control plane.

  • EKS / GKE: sidecar, optional admission (install_admission), and compute agent when registration vars are set.
  • AKS: sidecar in the blekline namespace today; add admission the same way as EKS when you extend the module.

Paths in the repo:

  • deploy/terraform/blekline-eks
  • deploy/terraform/blekline-gke
  • deploy/terraform/blekline-aks

Define cell profiles

A cell profile is a short-lived act envelope (vCPU, memory, max duration, isolation), not a full always-on sandbox. You size only the governed step, which helps control compute spend versus leaving a dev VM running for the whole agent session.

Attach a default on the workflow or pass cellProfileId on each run. POST /api/workspace/cell-profiles (session) or use the console; the API returns a generated profile_* id. System SKUs such as blekline-small ship for hosted Fast act.

create-cell-profile.json
{
  "name": "payments-act-small",
  "description": "Bounded act for payments tool path",
  "vcpu": 2,
  "memoryGib": 4,
  "storageGib": 8,
  "maxDurationMin": 30,
  "isolationProfile": "standard"
}

Start a cluster act run

Use the same act body as run-first, plus placement fields. Token needs runs:write.

run-with-profile.json
{
  "workflowId": "wf_your_workflow_id",
  "displayName": "Cluster act run",
  "executionTier": "act",
  "toolCall": true,
  "cellProfileId": "blekline-small",
  "clusterId": "ct_your_compute_target_id",
  "region": "eu-central-1"
}

Confirm compileHash on the run row, then read Runtime policy if you need to change the published guard set.