Oculis Collector

Applies to1.1.0DifficultyBeginner

Oculis Collector discovers the AI workloads running across your infrastructure and reports operational telemetry to Oculis.

It runs out-of-band: it observes the host it is installed on and sends data outbound to Oculis. It does not sit in the model request path, and no application or model traffic passes through it. If the Collector stops, your applications are unaffected — you lose visibility, not availability.

AI workloads  ─►  Oculis Collector  ─►  Oculis

Install Collector

Open Oculis, choose Add Collector, and copy the generated installation command for your platform. Treat the generated command as sensitive and run it only on the intended host.

On supported Linux, macOS, and Docker deployments, the installer can configure supported host instrumentation automatically, so no application-code changes or manual SDK installation are required. Windows Collector 1.1.0 provides workload discovery and host telemetry; application instrumentation is configured separately.

Linux / macOS

Prerequisites

• Root or sudo on the target host
• Python 3.10 or later — the installer detects it, and installs it if missing
• Outbound HTTPS on port 443 to api.oculis.selaware.ai

curl -sSL https://oculis.selaware.ai/install.sh \
  | sudo bash -s -- --deploy-token YOUR_DEPLOY_TOKEN

sudo goes on bash, not on curl.

The Collector is registered as a service, so it starts automatically at boot.

Windows

Run the PowerShell installer from an elevated prompt with your deploy token. The Collector is registered as a Windows service named OculisCollector.

Docker

Prerequisites

• A Linux Docker host running the applications you want visibility into
• Root or sudo
• Python 3.10 or later on the host — the installer detects it, and installs it if missing
• Outbound HTTPS on port 443 to api.oculis.selaware.ai

curl -sSL https://oculis.selaware.ai/install-docker.sh \
  | sudo bash -s -- --deploy-token YOUR_DEPLOY_TOKEN

The installer detects your platform, installs and configures the required Collector components, starts the Collector, and confirms it has connected to Oculis. You do not need a registry account or docker login.

The installer is safe to re-run. It detects an existing installation and reuses it rather than creating a second Collector.

To remove the Collector, re-run the same installer with --uninstall. Collector state is preserved by default so that reinstalling keeps the same Collector; add --purge-all to remove it.

Keep Collector state across upgrades

Removing the Collector’s stored state — including the Docker volume the installer creates — makes it register as a new Collector. Preserve it when upgrading or recreating the container.

Supported platforms

Platform	Architectures
Linux	64-bit Intel and ARM
macOS	64-bit Intel and ARM
Windows	64-bit Intel
Docker	64-bit Intel and ARM

Kubernetes and Helm deployment are not supported in Oculis Collector 1.1.0.

What Collector provides

Capability	Oculis Collector 1.1.0
AI workload discovery	Available
Host operational telemetry	Available
Telemetry from Oculis-instrumented applications	Available
Collector health and status reporting	Available
GPU telemetry	Not available
Kubernetes and Helm deployment	Not supported

GPU telemetry is not available in this release

Collector 1.1.0 does not report GPU utilization, memory, or temperature. If you are evaluating Oculis for GPU visibility, this release will not provide it.

Containerized AI workloads

Collector 1.1.0 can discover supported AI workloads running directly on the host and inside Docker containers on that host.

Full application telemetry from a containerized workload requires Oculis instrumentation inside that application’s environment. Discovery tells you the workload exists; instrumentation is what reports what it is doing.

Workload	Discovered	Full telemetry
AI application on the host	Yes	Yes, configured by the installer
AI application in a container	Yes	Requires instrumentation in its image
Non-AI containers	Not reported	—

Networking requirements

Direction	Destination	Port	Protocol
Outbound	api.oculis.selaware.ai	443	HTTPS

• No inbound access is required. The Collector exposes no endpoint and accepts no connections, so no firewall rules, port forwarding, or public exposure are needed.
• TLS certificate verification is enabled by default.
• Treat your deploy token as a secret, and restrict administrative access to the Collector host.

Check Collector health

Verification

Reports whether the Collector is running and connected to Oculis. Exit code 0 means healthy; 1 means it is stopped or unhealthy.

Linux / macOS

oculis-collector --status

Windows

oculis-collector --status

Docker

docker exec oculis-collector oculis-collector --status

A healthy Collector reports healthy for both the Collector itself and its connection to Oculis, along with when it last delivered data. A Collector that is stopped, hung, or unable to reach Oculis reports that explicitly and suggests the next step.

Add --json for machine-readable output.

To confirm which release is installed, run oculis-collector --version.

View Collector logs with your platform’s normal service or container logging tools — journalctl on Linux, Console on macOS, Event Viewer on Windows, or docker logs oculis-collector for a container deployment.

Troubleshooting

The Collector is not connected. Confirm outbound HTTPS on port 443 to api.oculis.selaware.ai is permitted, then re-run --status. The report names the last error it encountered.

No workloads are appearing. Workloads are reported as your applications run. Confirm the Collector is healthy, and that the applications you expect are Python, Node, Java, or .NET processes.

A discovered workload reports no telemetry. This is expected for a containerized application without Oculis instrumentation in its own environment — see Containerized AI workloads.

A backlog is building. Confirm connectivity to Oculis and make sure the host has sufficient free disk space.

The Docker container will not start. Check docker logs oculis-collector for the reason, and confirm Docker is running and the host meets the prerequisites above. Re-running the installer is safe and will repair a partial installation.

Versions

This documentation describes Oculis Collector 1.1.0 with Oculis SaaS 1.0.0.

The Collector and the Oculis platform are versioned separately. Where a version is relevant, it is named explicitly.