AI Coding Assistant Lingma:Coding with Lingma

Default keyboard shortcuts

Lingma provides ready-to-use keyboard shortcuts. The following table lists the most common ones:

Additionally, in the AI Chat panel, press Cmd+Enter (macOS/Linux) or Ctrl+Enter (Windows) to create a line break, and press Enter to submit your question to the model.

Custom keyboard shortcuts

IDE settings

Common configuration items

1. Disable auto-completion by file type.

   • If auto-completion creates interference when working with certain types of files, you can add those file extensions to the list. Multiple extensions should be separated by commas (for example: txt,md).

     IDE	Configuration
     JetBrains IDE
     VSCode

     Note

     You can disable auto-completion for specific file types to prevent automatic triggering of auto-completion. If you use the Alt+P keyboard shortcut to manually trigger auto-completion for the files, you can still use LLM-based code generation and completion.

2. Keep completion results when drop-down suggestions appear.

   • By default, when the IDE shows syntax-based drop-down completion suggestions, Lingma automatically stops showing large model completions to avoid visual conflicts.

   • If you want Lingma to always generate LLM completions, you can select this configuration option. As shown in the figure below, pressing the Tab key will accept the large model's generated results.

     IDE	Configuration
     JetBrains IDE
     VSCode

3. Control the maximum length of the generated code.

   • Lingma supports setting different code completion length parameters for both automatic triggers and manual triggers. It's recommended to set the manual trigger (default shortcut Alt+P) to generate slightly longer completions than automatic triggers.

     IDE	Configuration
     JetBrains IDE
     VSCode

     Note

     This configuration option only sets the maximum length that the model is allowed to generate. If the content generated by the model in a completion is naturally shorter, modifying this configuration cannot force the model to generate longer content.

Guide code completion with comments

When there are no guiding comments, the model can only predict upcoming code based on current context and similar code found in project references. When the model's predictions are inaccurate, you can add comments to guide the model towards implementing the desired code.

1. For example, in the following code, the model first predicts a CHAT_CONTEXT field, which is not what we expect.

2. In this case, add a comment to instruct the model to generate a historical record field. Then, the model generates the expected field and its corresponding data code.

Use descriptions to generate methods

1. You can generate entire methods based on comments by either Comment to Code in the editor or use the AI Chat panel. Since Lingma's AI Chat typically uses larger model parameters than the code completion model, directly ask your question in the AI Chat panel.

2. If you have specific requirements for the language or method signature (including method name, parameter types, return value types), describe these details in your prompt.

Save files and update indexes

1. Lingma's cross-file indexing is a crucial mechanism for preventing code hallucinations. By automatically identifying types and method definitions in the current context, the model can recognize members of types and method parameters from other files in the project. When first opening a new project, Lingma automatically creates the project's file index. Thereafter, each file save triggers an incremental index update for that file. However, due to IDE memory caching, when switching to another file right after editing one, the local index might not have updated to recognize newly added or modified content, causing the model to reason based on the original type structure. For example, in a code project, add a saleable property to the Pet object.

2. Then, switch to another file and try to let the LLM complete the code, but the model infers logic using another less relevant field.

3. To eliminate such information discrepancy, actively pressCtrl+S (Cmd+S for macOS/Linux) after editing the previous file, before continuing to edit other files. This ensures generated content correctly references the modified object structure.

Optimization solution for MyBatis scenarios

In addition to cross-file reference capabilities for mainstream programming language like Java, Python, and JavaScript, Lingma also supports automatically identifying the table structure types referenced by Mapper objects when writing MyBatis XML files. For example, when writing an insert statement, Lingma will use the TaxiOrder type information in the current project to ensure that each generated field is correct.

Clear context timely

In each chat flow, previous conversation content is automatically provided to the model as context. When you ask a question unrelated to the previous content, this extra information may interfere with the model's response.

Click New Chat at the top of the AI Chat panel to start a new chat flow, or use the /clear context command to existing context without starting a new chat flow.

View chat history

After creating a new chat flow, you can:

• Use the chat history feature to return to previous topics.

• Continue asking follow-up questions.

General questions

Two ways to ask questions about code:

• Paste the code into the chat box area.

• Select a piece of code in the code editor, and then ask questions about it, for example:

Built-in code tasks

Lingma has four built-in code tasks:

• Explain code

• Generate unit tests

• Generate code comments

• Optimize code

The Lingma LLM has been specially trained for these tasks. For unit test generation, use the required built-in task instead of selecting code and entering a generation method.

Lingma supports three methods for using the built-in code tasks. The most common method is to click the Lingma icon at the beginning of a method definition and select the task from the drop-down options.

Reference code in questions

When asking questions, if you select text or code in the editor, the selection will automatically be appended to your question in Markdown quote format. Therefore, when referring to the selected code in your prompt, use phrases like "the following code" or "this content". For example:

• Good example: Check whether the following code has a risk of an index out-of-bounds error.

• Bad example: Check whether the selected code has a risk of an index out-of-bounds error. (In this example, the model cannot identify the selected code.)

Add contexts to commands

Add supplementary information to commands to provide richer context for chat, resulting in more accurate responses.

Generate effective code through multi-turn chat

When conversing with large language models, providing richer context leads to more accurate and expected results. You can build upon previous chat turns to add contextual information for subsequent questions, allowing the generated results to better reflect the entire historical context. However, when previous historical information becomes interfering, you should clear context timely.

Based on the previous round, you can ask more in-depth follow-up questions.

Provide examples for the model

When you need the model to output in a specific format or follow certain rules, providing a reference example often works better than just text descriptions. For instance, when formatting program test results into a specific JSON structure, first open the file, select all content, then ask questions in the chat box.

Compare these two prompts, where the second one produces more consistent output:

• Prompt 1 (Less effective): Organize the file into a test report in the JSON format. Make sure that the results of each test case are organized into a JSON structure. Populate the name field with the test case name, the success field with the test case status, the duration field with the running duration in milliseconds, and the coverage field with the test coverage ratio. Make sure that the detail field is in the JSON format and includes the input and output fields for each test case.

• Prompt 2 (More effective): Organize the file into a test report in the JSON format.

  ...Report content omitted...

  Refer to the following example as an input report.

  [
    {
    "name": "If the page number exceeds the valid value range, the system is expected to return an empty list or prompt no-more-data",
    "duration": 3434,
    "coverage": 80,
    "detail": [
      {
      "input": "…",
      "output": "…"
      }
    ]
    }
  ]