Skip to content

Run with Lens Agents#

Lens Agents ships a built-in Prism Agent template that provisions Prism as a managed agent inside a project sandbox. The template pins the published image, exposes the web UI on a private port, mounts a persistent volume, and opts the agent into Nexus-managed LLM inference — so you don't supply a provider credential the way you do for a native or Docker install. A four-step wizard collects a name, optional Slack credentials, and any extra policies, then creates the credential, agent policy, and sandbox for you.

Use this path when you want Prism governed by Lens Agents: outbound traffic gated by a policy, Slack tokens injected at the boundary proxy rather than living in the agent's environment, and inference routed through the platform.

Create a Prism sandbox#

To provision Prism from the template:

  1. Open the project that will host the agent.
  2. In the Sandboxes section, click Create Sandbox, then select Prism Agent. The wizard opens on the Name step.
  3. Enter a name for the sandbox, then click Next.

  4. Optional. On the Credentials step, connect Slack to let the agent operate in a workspace over Socket Mode. Toggle Slack on and enter both tokens:

    Field Value
    Slack Bot Token The bot token, starting xoxb-.
    Slack App Token The app-level token for Socket Mode, starting xapp-.

    Leave Slack off to create the agent without it. Click Next.

  5. Optional. On the Policies step, attach existing project or organization policies, or add egress bundles from the policy templates. Each selection merges its allowed domains into the agent policy the wizard creates. Click Next.

  6. On the Review step, confirm the name, image, CPU and memory, Slack status, and attached policies, then click Create.

The wizard creates a Slack credential (when connected), an agent policy, and the sandbox. The agent starts on the pinned image and its web UI is reachable on the sandbox's private port. Open the sandbox from the Sandboxes table to view its state and interfaces.

What the template provisions#

The Prism Agent template seeds the sandbox with these defaults:

Setting Value
Image ghcr.io/lensapp/prism-agent:latest
Web UI port 3003, private
Volume /data, for persistent agent state
CPU / memory 1 core / 2Gi (adjustable in the wizard)
Managed inference Enabled — the agent reasons via the Nexus-managed LLM, so no provider credential is required

When you connect Slack, the template registers credential injections at the boundary proxy: the real bot and app tokens are applied to matching Slack API requests at the sandbox edge, while the agent's environment only ever holds a placeholder. A Slack method the template doesn't cover is denied before it leaves the sandbox and recorded in the audit trail.

Note

Connecting Slack through this template requires a Slack app configured for Socket Mode with both a bot token and an app-level token. See Slack for how the agent uses Slack once it's running.