Skip to main content
Target chat investigates one Kubernetes cluster or Linux VM with the target’s live capabilities. Each message creates a target-scoped run under the current user, workspace, access mode, provider, model, and tool policy.

​
Reference an exact tool or skill

Type / at a token boundary in the composer to open the target capability picker. It lists enabled tools and installed target skills that are available in the selected read-only or read-write mode. Selecting an item adds a removable reference chip. AcornOps submits the reference as a structured ID, resolves it against the current target, and freezes its qualified identity for the run. The chip label is not authorization. You can attach up to eight references. A write tool that is unavailable in the selected mode does not appear. Duplicate, disabled, stale, wrong-target, or unauthorized references fail explicitly so you can correct the message. References guide capability selection without widening the run:
  • tool and skill policy remains authoritative,
  • write-capable references still require read-write permission and approval,
  • Kubernetes RBAC or VM local policy can still reject an operation,
  • authorization is rechecked before execution.

​
Review compact and complete evidence

The run trace stores a bounded model-facing projection for each tool call. It identifies how the result was projected and whether content was omitted. Complete tool output is never inserted into the normal event stream. For eligible trusted results, View full redacted result loads a separate short-lived artifact. The viewer:
  • supports JSON and text,
  • shows artifact size and expiry,
  • renders at most 100,000 characters,
  • keeps the payload out of ordinary trace state until requested.
Artifacts are redacted before storage, limited to 2 MiB by default, and retained for seven days by default. Reading one requires current workspace membership and read_workspace_data, uses Cache-Control: no-store, and creates a workspace audit event. Missing, expired, inaccessible, and unavailable artifacts use the same unavailable response. An unavailable complete artifact does not invalidate the compact evidence already recorded for the run.

​
Export a completed answer

After an assistant turn completes, choose Create Document to export the persisted answer. The browser sends the source run and optional title; it does not ask the model to recreate the response. The resulting PDF or Markdown document is provenance-linked, authenticated, and retained for the deployment-configured document period. Document creation is a platform-native artifact operation, not an AgentK or AgentV tool, and it does not require a write-capable target run. A Workflow can also invoke the reviewed documents.create capability to create a document from approved Workflow evidence.

​
Collaborate without overwriting another investigation

The person who creates a target conversation can send follow-up messages while they retain run permission. Other workspace users with target read access can open the conversation and watch active events, but the composer remains read-only for them. When recent chat activity exists on the target, AcornOps warns before you start a separate conversation. Review the existing investigation first, especially when it used write-capable tools.

​
Retention and deletion

Conversation history, complete tool-result artifacts, and generated documents have separate retention policies. Deleting a run removes its retained complete-result artifacts. Expiry of a complete artifact does not delete the conversation or its compact evidence. Platform administrators configure these periods in the configuration reference.