Overview

The AI Runbook Assistant adds a Suggest runbook button to the Checklists card on the ticket detail page. When you press it, Core drafts an ordered list of steps the technician can follow — grounded in three signals pulled from your tenant:

• Knowledge base articles, matched by semantic similarity against the ticket title and description.
• Recently resolved tickets of the same ticket type or customer, so the collective memory of past work shows up.
• Authored checklist templates, so proven playbooks keep surfacing.

A runbook is a regular checklist — the same check, skip, note, and completion behavior you already use. It is distinguished from manual checklists only by a small "AI runbook" badge and two extra actions: Regenerate and Save as template.

---

Enabling the feature

The runbook assistant is on by default for every workspace — AI is included with your subscription, no API key setup required. The Suggest runbook button appears in the Checklists card header on every ticket detail page.

---

Day-to-day usage

Draft a runbook for a ticket

1. Open the ticket (/tickets/tkt-123).
2. Click Suggest runbook in the Checklists card header.
3. Core drafts a short list of steps, each with optional notes and estimated minutes.
4. Work the steps as a normal checklist: check off, skip, add notes, mark complete.

Regenerate when the ticket evolves

If the ticket description has changed materially since the runbook was drafted, click Regenerate on the runbook's ribbon. Core re-runs the retrieval pipeline and replaces the step list. Completed steps with matching titles are preserved (best-effort merge).

Save a runbook as a template

Once a runbook proves out, click Save as template on the ribbon. Name the template and Core creates a reusable checklist template that the whole team can pull on future tickets. The original runbook stays on the ticket as the audit trail.

---

How retrieval works

Three signals feed the draft:

1. Knowledge-base search — articles related to the ticket title and description. Articles for the same company rank higher. Other private company content is not included.
2. Prior resolved tickets — completed work with the same ticket type or company.
3. Checklist templates — workspace-wide and company-specific templates related to the ticket.

Steps cite their source (KB article id, prior ticket number, or template name) in the step note when retrieved material was used. Steps that fall outside retrieved context are produced by the model's general reasoning and are not cited.

---

Permissions

• procedures:read and procedures:execute — required to see and use the runbook controls.
• tickets:update — required to write the runbook back to the ticket.
• kb:read — required for KB retrieval.

---

Limitations

• The assistant draws only from your tenant's data; it will not reach across tenants or to the public web.
• Runbooks are drafts, not directives. Each step is editable and skippable, and sources are surfaced so technicians can verify before acting.
• The first call after regeneration or suggestion may take several seconds while context is assembled and the model streams its response.

Overview

The AI Runbook Assistant adds a Suggest runbook button to the Checklists card on the ticket detail page. When you press it, Core drafts an ordered list of steps the technician can follow — grounded in three signals pulled from your tenant:

• Knowledge base articles, matched by semantic similarity against the ticket title and description.
• Recently resolved tickets of the same ticket type or customer, so the collective memory of past work shows up.
• Authored checklist templates, so proven playbooks keep surfacing.

A runbook is a regular checklist — the same check, skip, note, and completion behavior you already use. It is distinguished from manual checklists only by a small "AI runbook" badge and two extra actions: Regenerate and Save as template.

---

Enabling the feature

The runbook assistant is on by default for every workspace — AI is included with your subscription, no API key setup required. The Suggest runbook button appears in the Checklists card header on every ticket detail page.

---

Day-to-day usage

Draft a runbook for a ticket

1. Open the ticket (/tickets/tkt-123).
2. Click Suggest runbook in the Checklists card header.
3. Core drafts a short list of steps, each with optional notes and estimated minutes.
4. Work the steps as a normal checklist: check off, skip, add notes, mark complete.

Regenerate when the ticket evolves

If the ticket description has changed materially since the runbook was drafted, click Regenerate on the runbook's ribbon. Core re-runs the retrieval pipeline and replaces the step list. Completed steps with matching titles are preserved (best-effort merge).

Save a runbook as a template

Once a runbook proves out, click Save as template on the ribbon. Name the template and Core creates a reusable checklist template that the whole team can pull on future tickets. The original runbook stays on the ticket as the audit trail.

---

How retrieval works

Three signals feed the draft:

1. Knowledge-base search — articles related to the ticket title and description. Articles for the same company rank higher. Other private company content is not included.
2. Prior resolved tickets — completed work with the same ticket type or company.
3. Checklist templates — workspace-wide and company-specific templates related to the ticket.

Steps cite their source (KB article id, prior ticket number, or template name) in the step note when retrieved material was used. Steps that fall outside retrieved context are produced by the model's general reasoning and are not cited.

---

Permissions

• procedures:read and procedures:execute — required to see and use the runbook controls.
• tickets:update — required to write the runbook back to the ticket.
• kb:read — required for KB retrieval.

---

Limitations

• The assistant draws only from your tenant's data; it will not reach across tenants or to the public web.
• Runbooks are drafts, not directives. Each step is editable and skippable, and sources are surfaced so technicians can verify before acting.
• The first call after regeneration or suggestion may take several seconds while context is assembled and the model streams its response.