Semantic Assets Documentation
A Unity Editor plugin that enables semantic search over your project assets using vector embeddings. Find assets by meaning, not just by filename.
Overview
Semantic Assets allows you to find assets by meaning, not just by filename. Simply describe what you're looking for in natural language, and the plugin will find the most relevant assets in your project.
- Natural Language Search Find assets by describing what you need, not by remembering exact filenames. Works across languages and asset packs with different naming conventions.
- Smart Visual Understanding Optionally searches based on what assets actually look like, not just their names. Handles poor naming and distinguishes similar assets.
- Flexible Asset Selection Configure which asset types should be searchable based on your project needs.
- Team Data Sharing Reuse indexed data across team members to avoid redundant processing.
Search Comparison
-
Unity built-in:
t:Prefab swordrequires knowing the filename contains "sword" -
Semantic search:
medieval weapon for knightfinds swords, axes, maces that match the concept -
Multilingual:
木製の椅子orholzstuhlfinds the same wooden chair assets
Requirements
-
Unity 6000.3
Development and tested on this version -
Qdrant Vector Database
Cloud or self-hosted — qdrant.tech -
OpenAI-compatible Embedding API
For generating vector embeddings from text -
Vision LLM API
(Optional) OpenAI-compatible Vision API for thumbnail-based descriptions
Configuration
Open Edit > Project Settings > Semantic Assets to configure the plugin.
Qdrant Settings
| Setting | Description |
|---|---|
| Base URL | Qdrant REST API endpoint (e.g., https://your-cluster.qdrant.io) |
| API Key | Qdrant API key for authentication |
Embedding Settings
| Setting | Description |
|---|---|
| API Key | API key for the embedding service |
| Base URL | (Advanced) API endpoint, change only if using a non-OpenAI provider |
| Model | (Advanced) Embedding model name (default: text-embedding-3-large) |
| Dimensions | (Advanced) Must match your model's output dimensions (default: 3072) |
Vision Settings
| Setting | Description |
|---|---|
| Use Embedding Credentials | When enabled, uses the same Base URL and API Key as Embedding settings |
| Base URL | Vision API endpoint (only when not using embedding credentials) |
| API Key | Vision API key (only when not using embedding credentials) |
| Model | (Advanced) Vision model name (default: gpt-4o-mini) |
Indexing Settings
Configure which file extensions to index and whether to use Vision enhancement.
| Extension | Enabled | Vision | Description |
|---|---|---|---|
| prefab | Yes | Yes | Unity Prefabs — uses Vision for rich descriptions |
| mat | Yes | No | Materials — indexed by path and file name only |
| png/jpg/tga | No | No | Images — disabled by default |
| fbx/obj | No | No | 3D models — disabled by default |
You can add or remove extensions as needed based on your project requirements.
General Settings
| Setting | Description |
|---|---|
| Share Collection | When enabled, team members share the same index data, avoiding redundant processing. When disabled, each user maintains their own independent index. |
| Search Result Limit | Maximum number of results returned per search (default: 20) |
Usage
Opening the Search Window
Open the search window via Window > Semantic Asset Search.
Performing a Search
- Enter your search query in the text field
- Press Enter or click Search
- Browse results sorted by relevance
- Click any result to locate and highlight the asset in the Project window
How It Works
- Indexing: On first run and periodically (every 60 seconds), the plugin scans your project for assets matching enabled extensions
- Embedding Generation: For each asset, generates a vector embedding from either the asset path (standard mode) or a Vision LLM description of the asset thumbnail (Vision mode)
- Vector Storage: Embeddings are stored in your Qdrant collection
- Search: Your query is converted to an embedding and matched against stored vectors using similarity search
Tips for Better Results
Prefabs and 3D models benefit significantly from Vision descriptions that capture visual details.
"Red medieval sword weapon" works better than just "sword" — more context leads to more accurate results.
Only enable extensions you actually need to search. This reduces indexing time and API costs.
Maintenance
Reset Collection
If you need to rebuild the index from scratch:
- Open Project Settings > Semantic Assets
- Click Reset Collection in the Maintenance section
- Confirm the action
This deletes the current Qdrant collection and rebuilds the index on the next sync cycle. This will consume API credits for re-indexing all assets.
Troubleshooting
This message appears when required settings are missing. Ensure you have configured:
- Qdrant Base URL and API Key
- Embedding API Key
- Vision Base URL and API Key (if not using embedding credentials)
- Check if assets with enabled extensions exist in your project
- Verify Qdrant connection (check Console for error messages)
- Try clicking Reset Collection to rebuild the index
- Verify Vision API credentials are correct
- Check Console for Vision API error messages
- Ensure Vision is enabled for the relevant extensions
Support
Need help? We're here for you.
- Documentation: semantic.1percentsync.games/docs.html
- Contact: hi@1percentsync.games