Znuny-LLM Add-on Announced: A Closer Look — and How OpenTicketAI Compares
Znuny just announced Znuny-LLM, a free, open-source add-on that brings large language model assistance directly into the Znuny agent and admin interface — running entirely on infrastructure you control. You can read the original announcement on the Znuny blog: "Znuny-LLM Add-on: Self-Hosted AI for Service Desks".
This is genuinely good news for the open-source ticketing ecosystem. Self-hosted, GDPR-friendly AI inside the helpdesk is exactly where the market should be heading. In this post we summarize what Znuny-LLM does, then compare it to OpenTicketAI — our open-source automation engine that takes a deliberately different, vendor-neutral approach — including the advantages and disadvantages of each.
Looking for the bigger picture? See our overview of AI-powered open source ticket systems covering OTOBO AI, Zammad AI, and OpenTicketAI.
What Znuny-LLM Does (Based on the Announcement)
According to Znuny's announcement, the add-on connects to a self-hosted LLM stack of your choice. Out of the box it talks to Ollama, but it also works with any OpenAI-compatible endpoint, including Azure OpenAI for teams that prefer a managed service inside their existing Microsoft tenant. There is no shared SaaS layer, no third-party gateway, and no telemetry leaving your environment. Znuny 7.3 or newer is required, and availability is planned starting July 8 via the Znuny open-source repository.
The feature set is broad and clearly designed to fit into everyday agent work:
| Feature | What it does |
|---|---|
| Automatic ticket classification | The LLM proposes a queue, type, service, and SLA from the values actually permitted for the ticket, shown in a sidebar widget with thumbs-up/thumbs-down voting. Confirmed votes are stored and reused as few-shot examples — improving results without retraining. |
| Multi-language ticket summaries | A short, factual 2–4 sentence summary per ticket, regenerated on new articles, stored per language and rendered in the agent's preferred language (EN, DE, FR, ES, IT, NL, PT). |
| Title rewriting for incoming email | Rewrites vague email subjects into a short, descriptive title in the customer's language. Phone and internal tickets are left untouched. |
| Dynamic field extraction by plain-language rule | Admins describe in plain language what to extract (invoice number, order ID, serial number) and the LLM writes it into the target dynamic field — no regex per format. |
| Compose Selection Rewrite | Agents highlight a passage while replying and the LLM polishes it, using the queue's answer prompt and respecting the detected language. |
| Vector-based template & FAQ suggestions | Most semantically similar templates and FAQ entries are ranked by cosine similarity over precomputed embeddings on the compose screen. |
| FAQ knowledge drafts from resolved tickets | An audit view lets admins generate FAQ drafts from good ticket conversations, with similar existing FAQs injected to avoid duplicates. |
Notably, Znuny-LLM was built with prompt injection in mind: incoming articles are scanned by a weighted multilingual pattern list before any LLM call, prompts separate trusted metadata from untrusted content with explicit boundary markers, and article bodies are truncated to a maximum length. These defenses ship as part of the default prompt design.
Our honest take: this is a strong, well-thought-out package. The native UI integration, the feedback-driven few-shot loop, and the security-first prompt design are all excellent. If you run Znuny 7.3+ and want AI inside that one system, it will be hard to beat for convenience.
OpenTicketAI: A Vendor-Neutral, Hybrid Approach
OpenTicketAI is our open-source automation engine and middleware that uses both trained ML classifiers and LLMs to classify, route, summarize, and enrich tickets — for Znuny, OTOBO, and Zammad alike. You can find it listed alongside the entire open-source ITSM ecosystem on Open ITSM Hub (openitsmhub.com), a directory where all the add-ons, plugins, and integrations for Znuny, OTOBO, OTRS, and Zammad are catalogued.
The philosophical difference is simple:
- Znuny-LLM is a native, LLM-first add-on for Znuny. It lives inside Znuny and leans on a general-purpose LLM for almost everything.
- OpenTicketAI is a system-agnostic middleware. It connects over the REST/Generic Interface and lets you choose the right tool per task — a small, fast, fine-tuned transformer (BERT/DistilBERT) for high-volume classification, and an LLM where generative quality matters (summaries, reply drafting).
How OpenTicketAI Works
- Connects via API to your existing system (Znuny, OTOBO, Zammad) — no forced version upgrade.
- Classifies queue, priority, type, and tags using either a trained classifier (fast, cheap, deterministic, no GPU needed at inference) or an LLM, depending on your needs.
- Confidence-based routing: high-confidence predictions auto-apply; uncertain ones land in a review queue for a human.
- Continuous improvement from agent feedback and periodic retraining.
- LLM features (summaries, response suggestions) via your own self-hosted or OpenAI-compatible models — fully on-premise.
Browse the full directory of open-source ticket-system add-ons — including OpenTicketAI — at openitsmhub.com.
Znuny-LLM vs. OpenTicketAI: Side-by-Side
| Aspect | Znuny-LLM | OpenTicketAI |
|---|---|---|
| Supported systems | Znuny only (7.3+) | Znuny, OTOBO, Zammad (and more) |
| Integration style | Native add-on inside the Znuny UI | Middleware over REST / Generic Interface |
| AI method | LLM-first (few-shot from feedback) | Hybrid: trained ML classifiers and LLMs |
| Classification | LLM proposes queue/type/service/SLA | Fine-tuned transformer or LLM, confidence thresholds |
| Hardware | Needs an LLM stack (Ollama/GPU or Azure OpenAI) | Trained classifier runs on CPU; LLM optional |
| Generative features | Summaries, title rewrite, compose rewrite, FAQ drafts | Summaries & reply suggestions via your LLM |
| Data residency | Fully self-hosted, no telemetry | Fully on-premise / self-hosted |
| Prompt-injection defense | Built-in, multilingual scanning | Configurable; trained classifiers aren't prompt-injectable |
| Best for | Teams committed to Znuny wanting deep native AI | Teams wanting one AI layer across multiple systems |
Advantages of Znuny-LLM
- ✅ Deep native integration — sidebar widgets, compose tools, and audit views feel like part of Znuny.
- ✅ No training required — works out of the box and improves via thumbs-up/down few-shot feedback.
- ✅ Broad generative feature set — classification, summaries, title rewriting, field extraction, FAQ drafting in one package.
- ✅ Security-first design — prompt-injection scanning and prompt boundaries by default.
- ✅ Free and open-source, fully self-hosted.
Disadvantages of Znuny-LLM
- ⚠️ Locked to Znuny — no help if you also run OTOBO, Zammad, OTRS, or are mid-migration.
- ⚠️ Requires Znuny 7.3+ — older instances must upgrade first.
- ⚠️ LLM-only classification — every classification is an LLM call, which means latency, GPU/infra cost, and non-determinism at high ticket volumes, where a tiny trained classifier would be faster, cheaper, and more consistent.
- ⚠️ Needs an LLM stack — you must operate Ollama/a GPU or pay for Azure OpenAI.
Advantages of OpenTicketAI
- ✅ Vendor-neutral — one automation layer across Znuny, OTOBO, and Zammad; ideal for mixed estates and migrations.
- ✅ Hybrid AI — use a fast, cheap, deterministic trained classifier for high-volume routing and reserve the LLM for generative tasks.
- ✅ No forced upgrade — connects to your current version over the API.
- ✅ Confidence-based routing with a human-in-the-loop review queue.
- ✅ Lower running cost at scale — classification doesn't require a GPU.
- ✅ Listed openly on Open ITSM Hub next to the whole ecosystem.
Disadvantages of OpenTicketAI
- ⚠️ Less "native" — it's a separate middleware service rather than UI widgets baked into one product.
- ⚠️ Setup effort — you deploy and connect the engine yourself.
- ⚠️ Trained-classifier path needs data — the ML route benefits from historical tickets (the LLM route does not, but then you're closer to Znuny-LLM's model).
- ⚠️ You operate it — more flexibility means more configuration than a single click-to-install package.
Which Should You Choose?
There's no single winner — it depends on your landscape:
- You're all-in on Znuny 7.3+ and want the simplest path to in-UI AI? Znuny-LLM is an excellent, well-engineered choice. Use it.
- You run more than one system (e.g., OTOBO and Zammad), are migrating, sit on an older Znuny, or process high ticket volumes where LLM-per-ticket cost and latency matter? OpenTicketAI's hybrid, vendor-neutral approach will serve you better.
- You want both worlds? Many teams use a trained classifier for routing (cheap, instant, deterministic) and an LLM for summaries and reply drafting — exactly the split OpenTicketAI is built around.
The Bigger Picture
The fact that Znuny is shipping a first-party, self-hosted LLM add-on validates what we've believed all along: AI in open-source ticketing should be local, transparent, and under your control. More options here is a win for everyone, and we welcome Znuny-LLM to the ecosystem.
If you want an AI layer that isn't tied to a single product — and that lets you mix a lean trained classifier with an LLM — take a look at OpenTicketAI, explore the full add-on directory at openitsmhub.com, or contact us to discuss the right setup for your helpdesk.
Frequently Asked Questions
Is Znuny-LLM free?
Yes. According to Znuny's announcement, Znuny-LLM is a free, open-source add-on, available via the Znuny open-source repository starting July 8 (Znuny 7.3+ required).
Does Znuny-LLM send data to the cloud?
No shared SaaS layer is involved. It connects to a self-hosted LLM (e.g., Ollama) or any OpenAI-compatible endpoint you choose — including Azure OpenAI inside your own tenant — so you control where ticket data goes.
How is OpenTicketAI different from Znuny-LLM?
OpenTicketAI is vendor-neutral middleware that works with Znuny, OTOBO, and Zammad, and it can use either a fast trained ML classifier or an LLM. Znuny-LLM is a native, LLM-first add-on for Znuny only.
Can I use OpenTicketAI with Znuny too?
Yes — OpenTicketAI connects to Znuny (as well as OTOBO and Zammad) over the REST/Generic Interface, without requiring you to upgrade to Znuny 7.3.
Where can I find all Znuny, OTOBO, and Zammad add-ons?
Open ITSM Hub (openitsmhub.com) is a directory cataloguing add-ons, plugins, skins, and integrations across Znuny, OTOBO, OTRS, and Zammad — including OpenTicketAI.