Connectors

Let your whole team run Argo Workflows.

One helm chart. You install it once. After that, anyone with a dagweave account can run Argo Workflows, not just you.

The problem

You already run Argo Workflows. Most of your team still cannot use it.

You run the cluster and you paid for the expertise. Using it means hand-writing Workflow YAML. That keeps it with the handful of people who can.

Everyone else works around that. They do the job by hand, or they reach for a Shadow IT tool that nobody on your team supports or maintains.

dagweave lets your whole team build workflows. The connector dispatches them to the cluster you already paid for.

Enrollment

One connector per cluster

dagweave generates the command, enrollment token included. The operator installs the chart once per cluster, then names each Argo Workflows instance the connector should manage.

Your cluster credentials stay in your cluster. dagweave never holds a kubeconfig.
The connector opens every connection outbound. Nothing needs an inbound port.

the install command dagweave generates bash

helm install dagweave-connector \
  oci://ghcr.io/dagweave/charts/connector \
  --namespace connector-system --create-namespace \
  --set connectorConfig.dagweaveEndpoint=<endpoint> \
  --set-string enrollment.token=<one-time token>

High availability

Survives a node drain

It ships production-ready, and there is nothing to tune.

It still comes up on a single-node cluster.

values.yaml, unmodified yaml

replicaCount: 2
leaderElection:
  enabled: true
podDisruptionBudget:
  enabled: true
  maxUnavailable: 1

What lands in the cluster

The log collector

A step fails and you want its log. A stock install brings the collector. Point it at S3, or let dagweave store your logs, and each one lands on the step.

Logs go to the console by default. Send them to your own S3, Loki or HTTP endpoint and dagweave never sees them.
Or let dagweave store them for you. With S3 or dagweave, each log lands on the step that produced it.
It only reads pods Argo Workflows labelled, nothing else on the node.

helm template, summarised text

Deployment    connector        x2, non-root, read-only rootfs
Service       ClusterIP        health + metrics only
PDB           maxUnavailable: 1
DaemonSet     log collector    every node except control-plane
                               root, tolerates every taint
                               /var/log                     ro
                               /var/lib/docker/containers   ro
                               its own state dir            rw
                               omitted when features.logs=false

Dispatch

Your workflows run on your cluster

The connector holds an outbound stream open. dagweave sends the compiled manifest over it, your Argo Workflows instance runs the workflow, and status comes back on the same stream. Nothing runs in dagweave.

We send the compiled Argo Workflows manifest, the same file your git repo gets.

enroll and dispatch text

helm install  ->  operator enrolls the connector, once
dispatch      ->  Workflow / WorkflowTemplate / CronWorkflow
report        ->  phase, per-step status, log refs, artifact URLs

Roles

Authors trigger runs without cluster access

The operator holds the cluster credentials and installs the connector. After that, the people who build workflows dispatch runs on a dagweave permission alone.

No kubeconfig, no cluster RBAC.
The operator keeps control of what the connector can reach, in the cluster.
When a dispatch does not arrive, the connector's own logs are in dagweave. Nobody reaches for kubectl to find out why.

Connect the cluster you already have.

dagweave is in early access. Leave an email and we will bring you on as we open it up.