Kai Options: Completions

Go Up to Kai Options

Tools > Options > Kai > Completions

Here, you can configure AI code-completion providers for Kai.

Kai supports multiple completion engines, including cloud-based providers and local models. You must sign in or configure the selected provider before using AI completions.

Completions Options

The following options are available in Completions:

Item	Description
General	Contains general AI completion settings. Enable auto completions Enables or disables AI code completions in the Code Editor. Show as Selects how AI completions are displayed in the editor. Ghost text Displays inline completion suggestions directly in the editor. CodeInsight symbol Displays completion suggestions using a symbol indicator. Active model Selects the AI provider used for code completions. GitHub Copilot Uses GitHub Copilot for code completions. Local model Uses a locally installed LLM through supported local providers.
Model settings	Contains settings for the selected AI provider. GitHub Copilot Displays settings for GitHub Copilot. Since this is an online model, you must first sign in. Status Displays the current connection or authentication status. Message Displays provider-related status or authentication messages. Sign in Opens the authentication process on a browser. Sign out Signs out from the selected provider. Local model Displays settings for the selected local model provider. Backend Selects the backend used for the chosen local model provider. Default backend Uses the default local model backend. Other (e.g. LM Studio) Uses another compatible local model backend, such as LM Studio. URL Specifies the URL address used to connect to the selected local model provider. Test connection Button to test the connection to the selected local model provider. Name Specifies the local model name. Context size Specifies the context size used by the local model.

Kai uses a Bring Your Own Key (BYOK) model. Depending on the selected provider, you may need a valid account or API key.

FIM Tokens Tab

FIM (Fill-In-the-Middle) tokens are special control tokens that tell a language model to complete text between a given prefix and suffix, rather than simply continuing from the end of a prompt. They are particularly important for code completion in IDEs, where the model needs to insert code at the cursor while preserving the code that follows. Configuring these token parameters allows user integration with additional local AI engines.

Since different model families use different special tokens, Kai now lets you choose predefined options or manually customize the FIM tokens.

Note:
To determine if a local LLM supports FIM, check the model information on the backend model, such as Hugging Face or the vendor’s documentation. It often explicitly states “Supports Fill-in-the-Middle (FIM).”

Most popular local coding models with FIM support include:

• StarCoder
• CodeLlama
• DeepSeek-Coder
• Qwen2.5-Coder
• Codestral
• Replit Code models

See Also

• Kai Options: Chat
• Kai Options: MCP Servers
• Kai Features