Skip to main content

​
Session

img Before sending a message, make sure the Asset Database and Unity Editor Connection in the upper-left corner are both working properly.
  • Multimodal input: the input box supports image upload and parsing. You can paste with CTRL+V.
  • Create a new session: click the + icon in the upper-left corner or use the shortcut CTRL+N (the shortcut can be customized on the Settings page).
  • Execution permission control: located in the lower-right corner. When auto mode is enabled, the Agent’s tool calls will run without approval. When it is disabled, any modification by the Agent will request permission from the user. This mode is safer, but it requires manual confirmation.
  • Model switching: located in the lower-right corner and used to switch the AI model for the current session.
  • Send message: the default shortcut is ENTER. You can change it to CTRL+ENTER on the Settings page.
  • Session list management: click > next to an item in the list on the left to expand the detailed execution process of a Subagent. List items support a right-click menu for actions such as archive and delete, and they support batch multi-selection while holding SHIFT.

​
Collaboration

img
  • Version control interaction: the terminal area at the bottom supports directly entering standard Git commands or giving instructions to AI in natural language.
  • Context menu: right-click a commit record or a modified file to open the extended function menu. The feature matrix is still evolving, and feedback is welcome.
  • Change preview and parsing: click the file list on the right, including the Commit, Stash, and Unstaged/Staged areas, to preview diffs. The system natively supports visual parsing of Unity YAML asset files such as Scene, Prefab, and ScriptableObject.
  • Conflict resolution: when merge conflicts occur, the system supports context-based semantic merging at the field level.
  • Locus-related file markers: files marked as related to Locus are the Agent’s own knowledge or configuration files. If you do not want them included in project version control, we recommend manually adding their paths to .gitignore

​
Knowledge

img

​
Knowledge Categories

The knowledge base is the core source from which the Agent obtains project context. The current knowledge system is divided into the following four categories:
  • Design: records project design requirements and factual baselines. AI can automatically update this document during a conversation according to business requirements, with user approval. Users can also edit and maintain it manually at any time.
  • Memory: dynamic context automatically maintained by AI. The system provides three built-in modules by default: user preferences, a mistake notebook, and project understanding. Users can create documents and define exclusive maintenance rules. AI will iterate on them automatically according to those rules.
  • Skill: used to solidify standardized high-frequency workflows. Users can describe a specific process in natural language and instruct the Agent to follow this Skill when it encounters a similar scenario.
  • Reference: an external read-only document library. It currently supports importing local documentation such as the Unity API Reference or connecting to external collaboration platforms such as Feishu Docs.

​
Retrieval Strategy

The system supports the following two document retrieval mechanisms:
  • Lexical indexing (enabled by default): exact text matching based on keyword extraction.
  • Semantic indexing (enable it in the retrieval settings below): based on text vectorization. The system retrieves semantically similar paragraphs from the document library according to semantic fragments produced on the AI side. It is suitable for fuzzy document queries.
img

​
Settings

The Settings page is used to manage global system settings.

​
Model Configuration

It supports customizing model parameters, configuring third-party API endpoints, and managing the sign-in and authorization status of subscription accounts. img