TriageAgent: a fully local AI that triages support tickets for my dev team(HALF BAKED)

Описание и применение

TriageAgent автоматически группирует сообщения по ID тикета, создает локальный индекс и, при поступлении нового тикета, либо генерирует внутреннее сообщение, если найден похожий решенный тикет, либо перенаправляет его разработчику, наиболее подходящему по профилю. Все операции выполняются локально без выхода данных в облако.

Задача

Dev teams spend a lot of time triaging support tickets, reading a wall of messages to determine which tickets are resolved, which need engineering, and routing them.

Результат

TriageAgent reduces manual triage effort by automating note generation for resolved duplicates and routing new tickets to the most relevant developer, while keeping all data local and free of cloud dependencies.

Рабочий процесс

• Parse free‑form chat log and group messages by ticket ID
• Embed each thread with nomic-embed-text and store in a local vector index
• Create a profile embedding for each developer from their past messages
• When a new ticket arrives, embed it and search the index
• If a similar resolved thread exceeds a similarity threshold, generate a short internal note using llama3.2:1b
• Otherwise route the ticket to the developer whose profile is most similar
• Print reports: open tickets per dev, critical cases, full case list

Доказательства

My dev team's biggest time-sink isn't writing code — it's triaging support tickets.

Источник

TriageAgent does that automatically. You paste in our support chat log — no strict format, just messages that mention a ticket ID like #123 — and it: groups messages into ticket threads by their ticket ID, builds a local index of every past thread and of each developer's past work, for a new ticket, either writes a short internal note to the support team (when a similar resolved thread exists) or routes it to the right dev based on whose past responses it most resembles, and prints case-based reports.

Источник

No cloud, no API keys, nothing leaves the machine.

Источник