ai-triage

Enterprise Part of the commercial Kumiko Enterprise offering — not in the open framework.

Query handler (ai-triage:query:triage) that classifies an inbound message (subject, sender, body) against a caller-supplied list of action types and returns a category and proposed next actions. The system prompt is editable per-tenant via the prompt-store or a config-key so operators can tune classification rules without a deploy; the caller applies the result in a separate action.

How it fits

What this feature needs to run (Requires, top) and the write commands it provides (Provides, bottom).

flowchart TB
  n_ai_triage["ai-triage"]
  subgraph how_reqs["Requires"]
    n_ai_foundation["ai-foundation"]
    n_config["config"]
    n_prompt_store["prompt-store"]
  end
  n_ai_foundation --> n_ai_triage
  n_config --> n_ai_triage
  n_prompt_store --> n_ai_triage

Dependencies

• Requires: ai-foundation, config, prompt-store
• Activation: always on (not toggleable)

Configuration

Per-tenant config keys, set via the tenant-admin UI or a seed. 🔒 = encrypted at rest.

Key	Type	Default	Scope	Who can write	Who can read
triage-system-prompt	text	You triage an inbound message for a business application.\n\n1. Classify it into a short lowercase category (e.g. "maintenance", "billing", "complaint").\n2. Propose 0–7 next actions, choosing ONLY action types from the provided catalogue — never invent a type.\n For each: a confidence in [0,1] and a one-sentence rationale.\n\nIf the message is too ambiguous to classify, call clarify_request instead of triage_message.\nCall exactly one tool.	tenant	TenantAdmin, SystemAdmin	TenantAdmin, SystemAdmin