Automated Troubleshooting for Installers via Selora Connect

Overview

Selora already gives installers a way to reach a customer’s Home Assistant installation: secure, on-request remote access provides temporary, time-boxed access once the customer approves the request. The workflow beyond that hand-off is still manual — and it requires the installer to be a skilled Home Assistant troubleshooter: reading logs, inspecting entities and integrations, and identifying what broke.

This feature makes the support session itself automatic. Once a temporary access window is active, Selora Connect uses AI over a secure connection to the installation’s MCP server to gather system state, correlate it with the reported symptom, produce a diagnosis, and suggest a concrete fix — so the installer can resolve the issue without diagnosing it from scratch.

Customer value

• Faster support: The diagnosis is produced while the installer is still reading the customer’s message, not after an hour of manual investigation.
• Lower skill floor: Installers who are not full Home Assistant experts can still handle most issues — Selora explains what is wrong and how to fix it in plain terms.
• Consistent outcomes: Every support session follows the same diagnosis-first approach, so resolutions are repeatable and reviewable.
• Less disruption: The AI collects evidence through the existing secure connection; the installer is not asked to screenshot logs or run commands.
• Customer confidence: Customers see their installer close issues quickly, with explanations they can understand.

Scope

• Symptom intake: The installer describes the customer’s issue in natural language in Selora Connect; Selora AI structures it into a troubleshooting goal.
• Secure diagnostics channel: Connection from Selora Connect to the installation’s MCP server, scoped to the active temporary-access window and read-centric (state, logs, integrations, entity registry).
• AI diagnosis engine: Correlates the symptom with live system state to identify likely root causes, ranked with supporting evidence.
• Fix suggestions: Concrete, step-by-step remediation — entity or integration settings, configuration changes, or known-workaround references — with the relevant evidence cited.
• Installer confirmation: Selora proposes; the installer reviews and approves any action before it is executed. Nothing is applied automatically to the customer’s installation.
• Session record: Each troubleshooting session (symptom, evidence, diagnosis, actions) is stored for the installer and searchable across properties.

Architecture

1. Access gate: Troubleshooting is only available while a valid temporary-access window from the secure remote access feature is open.
2. MCP bridge: Selora Connect opens a secure, authenticated channel to the device MCP server inside the customer’s installation; the channel inherits the access window’s expiry.
3. Evidence collector: The AI agent queries system state, logs, and integration health through MCP tools, tailored to the reported symptom.
4. Diagnosis: Selora AI analyzes the collected evidence and produces ranked root-cause hypotheses with citations to the evidence.
5. Remediation plan: A step-by-step fix is generated; the installer approves each step that writes to the installation.
6. Audit trail: The full session is recorded and linked to the customer property and the original support request.

Open questions

• Permission boundary: Which MCP tools may the AI invoke without installer approval — read-only queries, or limited writes?
• Multi-cause issues: How to present complex, cascading failures (e.g., a network event that breaks three integrations) without overwhelming the installer?
• Customer notification: Should the customer be informed, in real time, that a diagnostic session is running on their installation?