Local AI Models

What local AI is

ClashWise generates clash titles with an AI model that runs on your own computer — a private AI engine installed alongside the plugin. That means:

Privacy — your element names, file names, and clash data never leave your PC during naming.
Works offline — after a one-time model download, naming works without an internet connection.
No per-title limits — local naming is unlimited on a paid plan; there are no naming credits or per-clash charges.
Stability — the AI runs as a separate process, so even if it fails it cannot crash your Navisworks session.

Local AI powers clash-title generation (the Enhance tool) only. Other AI features — the Wise assistant and cloud clash analysis — run in the ClashWise cloud.

Who can use it

The AI Models manager requires sign-in and a Starter subscription or higher (any Clash Management plan includes it).

Where to find it

Ribbon: ClashWiseAI tab → Clashes panel → Local AI (Alt+1), or
From Enhance: Settings tab → Manage models...

The first time you click Enhance with no model downloaded, a message titled "Local AI setup required" appears and the AI Models window opens automatically so you can set one up.

The AI Models window

The window is headed AI Models — "Download a model to run Clash Title Updates locally without contacting the ClashWise server."

The model catalog

The list shows around ten open-source models, from roughly 1 GB to 11 GB. Use the Filter by provider: dropdown (All providers, Alibaba, Google, Meta, Microsoft, Mistral AI, OpenAI) to narrow it.
Each row shows the model's name, a short description, its download size, its status (Active, Downloaded, Not downloaded, or a download progress percentage), the date it was downloaded, the languages it supports, and Speed and Accuracy bars.
The recommended starter model — Qwen 2.5 1.5B, about 1 GB, supporting all 11 output languages — carries a green Recommended badge.

Downloading, activating, and deleting

Each model row has three buttons:

Download — fetches the model to your machine. The footer shows progress, and you can abort with Cancel Download.
Use — makes a downloaded model the active one. The active model is what Enhance uses, and it is shown in the Enhance window's AI Model panel.
Delete — removes the model file from your machine to free disk space. You can re-download it any time.

Use Refresh in the footer to reload the catalog, and Close when you are done.

Choosing a model: speed vs. accuracy

Smaller models (like the recommended ~1 GB starter) are fast and run comfortably on ordinary laptops. For most clash-naming work they are the right choice.
Larger models produce more refined titles but are slower and need more memory. The model descriptions include informal hardware guidance — for example, mid-size models are recommended for desktops rather than laptops, and the largest need workstation-class memory.
Every model's supported languages are listed on its row — if you name clashes in a language other than English, pick a model that supports it (the recommended model supports all 11).

Testing the active model

The Test the active model panel lets you try a real, one-off title generation without touching your clash data:

Enter two element names in Item 1 and Item 2 (defaults: 150mm Pipe and Steel Beam W12x26).
Pick a language.
Click Generate test title. The generated title appears along with how many milliseconds it took — a quick way to gauge the model's speed on your hardware.

If the test fails or returns an empty response, try re-downloading the model or choosing a different one.

Offline behavior

Downloading a model requires an internet connection, as does your first sign-in.
After that, Enhance works fully offline. Your subscription is re-verified periodically: after up to 14 days without a connection, you will be asked to reconnect once before continuing.

Where models are stored

Models are stored in your Windows user profile (one file per model). You do not need to manage the files directly — use the Delete button in the AI Models window to remove a model and reclaim disk space.

Troubleshooting

"Local AI setup required" — no model is downloaded or active yet. Download one and click Use.
A message that the local AI engine is not installed — part of the plugin installation is missing. Reinstall the ClashWise plugin; if the message persists, contact support at support@clashwise.ai.
Download failed — check your connection and click Download again, or use Refresh to reload the catalog.