The AI Code Completion
A VS Code extension that provides AI-powered inline code completions using the DeepSeek FIM API (Fill-in-the-Middle).
Features
- Inline completions — suggestions appear as ghost text as you type (using VS Code's
InlineCompletionItemProvider)
- Context-aware — automatically includes other open tabs as surrounding context for better completions
- Caching — repeated suggestions for the same prefix/suffix are served instantly within a 5-second TTL
- Debounced requests — avoids excessive API calls while you're still typing (configurable delay)
- Status bar indicator — shows idle/loading/no-key states and provides quick access to settings/API key configuration
- Token limit protection — truncates prompts to stay within model limits
Requirements
Installation
- Install the extension from the VS Code Marketplace (or build from source — see below)
- Open VS Code
- Set your API key via the command palette:
The AI Code Completion: Set API Key
- Completions will start appearing automatically as you type
Configuration
| Setting |
Default |
Description |
theAICodeCompletion.enabled |
true |
Enable/disable completions |
theAICodeCompletion.model |
deepseek-v4-flash |
Model ID for FIM completions |
theAICodeCompletion.maxTokens |
256 |
Maximum tokens per completion |
theAICodeCompletion.debounceMs |
500 |
Debounce delay in milliseconds |
theAICodeCompletion.maxContextFiles |
5 |
Number of open tabs to use as context (0 = disable) |
theAICodeCompletion.temperature |
0.1 |
Sampling temperature |
theAICodeCompletion.timeoutMs |
10000 |
Request timeout in milliseconds |
theAICodeCompletion.debug |
true |
Enable debug output channel |
theAICodeCompletion.stop |
["\\n\\n\\n"] |
Stop sequences |
Commands
The AI Code Completion: Set API Key — prompts for and stores your DeepSeek API keyThe AI Code Completion: Show Output — opens the debug output channel
Development
# Install dependencies
npm install
# Compile
npm run compile
# Watch mode
npm run watch
Build from source
npm install
npm run compile
The compiled output goes to out/extension.js.
How it works
- As you type, the extension captures the text before and after the cursor (prefix and suffix)
- Open editor tabs are collected as additional context (skipping binary files and files larger than 20 KB)
- A request is sent to the DeepSeek FIM API with the prompt, suffix, and context
- The returned completion is displayed as an inline ghost text suggestion
- Accepting a suggestion triggers a short guard period to prevent immediate re-triggering
License
MIT
| |
