Your first config

If you followed the quickstart, you have an agent connected and reporting healthy. The next thing worth doing is replacing the on-disk config it started with — and watching the live reload happen.

This walkthrough takes about five minutes.

1. Open the channel's Configurations page

In the UI, navigate to your channel and click Configurations. New channels show an empty state with three step-cards (create → version → deploy); click New config.

2. Author the config

Pick the agent type that matches your fleet — otelcol, fluent-bit, vector, telegraf, or refinery. Give the configuration a name (e.g. edge-collector) and paste in the agent's config body.

For an OpenTelemetry Collector starter:

receivers:
  otlp:
    protocols:
      grpc:
        endpoint: 0.0.0.0:4317

processors:
  batch:
    timeout: 5s
    send_batch_size: 1000

exporters:
  debug:
    verbosity: detailed

service:
  pipelines:
    traces:
      receivers: [otlp]
      processors: [batch]
      exporters: [debug]

Save. Ampbase validates the config against the agent's schema before persisting, so a typo here gets flagged immediately rather than after deploy.

3. Deploy this version

Saving creates version 1. The version is current but not yet deployed — agents are still on whatever they booted with. Click Deploy this version.

The deploy sheet offers two ways to ship it. With one agent connected, pick Everyone now and click Set default: the version becomes the channel default, and every connected agent picks it up on its next OpAMP message.

On a real fleet, leave Staged rollout selected instead. Ampbase ships the version to a first stage — 10, 50, 100 percent by default, or through the values of a label such as env — and widens to the next stage only once the agents on it report the version applied and healthy. When a version needs build components, it can only judge agents whose supervisor reports theirs; see Rollouts, under What automatic judgement needs from your fleet.

4. Watch the agent reload

Switch back to Agents. Within a second or two, your agent's row updates:

• config_hash changes to the new version's hash.
• The agent's status stays HEALTHY once the supervisor has written the file, sent SIGHUP, and the post-apply health check passes.
• If the agent fails its health check, the supervisor reverts to the previous config and the status flips to DEGRADED with the reason shown.

You can verify on the host too. The supervisor logs:

INFO  config_apply  hash=01HJK… bytes=412  result=applied
INFO  health_check  status=healthy

5. Iterate

Edit the configuration again, save — that's version 2. The current pointer moves; the deployed pointer doesn't. Diff the two versions from the UI, deploy when you're ready.

To roll back: open the version list, click Deploy this version on an earlier version, and choose Everyone now. Same primitive, opposite direction. (A staged rollout that is still in progress has its own Roll back button.)

Going further

• Rollouts — ship the next version in stages, each judged before the next.
• Routing — keep a region or a team on its own version, or A/B two configs.
• Configurations — versioning, bundles for multi-file agents, validation rules.
• Webhooks — notify your incident channel when a deploy lands.