DeepSeek Autocomplete
AI-powered code completions for VS Code using DeepSeek's Fill-in-the-Middle API.
Fast, context-aware inline suggestions that work like GitHub Copilot — powered by DeepSeek.
Features
- Ghost text completions appear inline as you type
- Fill-in-the-Middle understands both prefix (before cursor) and suffix (after cursor) context
- Streaming — suggestions appear progressively, no waiting for full response
- Multi-line detection — automatically offers block completions after
{, :, def, function, etc.
- Multiple alternatives —
Ctrl+Space shows 3 variations at different temperatures
- LRU cache — repeated patterns complete instantly from memory
- Smart filtering — skips sensitive files (
.env, secrets, credentials), configurable glob patterns
- Usage statistics — tracks shown/accepted completions, tokens used, estimated cost
- 14+ languages with tuned stop tokens per language
Requirements
Quick Start
1. Install the extension
Search for "DeepSeek Autocomplete" in the VS Code marketplace or install from VSIX.
2. Set your API key
Ctrl+Shift+P → DeepSeek Autocomplete: Set API Key
Paste your key — it's stored securely in your OS keychain, not in settings files.
3. Start typing
Open any supported file. Ghost text suggestions appear inline as you pause while typing.
| Key |
Action |
Tab |
Accept completion |
Escape |
Dismiss |
Ctrl+Shift+. |
Toggle on/off |
Ctrl+Right |
Accept word-by-word |
Ctrl+Space |
Show 3 alternative completions |
Configuration
All settings are under deepseekFim.* in VS Code settings (Ctrl+, → search "deepseek").
Essential
| Setting |
Default |
Description |
deepseekFim.model |
deepseek-v4-flash |
Model: v4-flash (fast) or v4-pro (quality) |
deepseekFim.enabled |
true |
Master toggle |
deepseekFim.triggerMode |
automatic |
automatic = while typing, manual = only on Ctrl+Space |
| Setting |
Default |
Description |
deepseekFim.debounceMs |
300 |
Delay before requesting (lower = faster but more API calls) |
deepseekFim.maxTokens |
256 |
Max tokens per completion |
deepseekFim.temperature |
0.0 |
0 = deterministic, higher = more creative |
deepseekFim.streamingTimeout |
500 |
Max ms to wait for streaming before showing partial result |
deepseekFim.cacheSize |
500 |
LRU cache entries (0 = disabled) |
Context
| Setting |
Default |
Description |
deepseekFim.maxPrefixChars |
6000 |
Max characters before cursor sent as prompt |
deepseekFim.maxSuffixChars |
2000 |
Max characters after cursor sent as suffix |
deepseekFim.maxPrefixLines |
100 |
Line-based cap (applied after char limit) |
deepseekFim.maxSuffixLines |
50 |
Line-based cap (applied after char limit) |
Advanced
| Setting |
Default |
Description |
deepseekFim.baseUrl |
https://api.deepseek.com/beta |
Change for custom/enterprise deployments |
deepseekFim.timeoutMs |
10000 |
HTTP request timeout |
deepseekFim.multilineCompletions |
auto |
auto, always, or never |
deepseekFim.stopSequences |
[] |
Custom stop sequences appended to language defaults |
deepseekFim.disableInFiles |
[] |
Glob patterns to skip (e.g. **/*.min.js) |
deepseekFim.debug |
false |
Verbose logging to output channel |
deepseekFim.snippetSupport |
false |
Enable tab-stop placeholders ($1, ${1:name}) |
Commands
| Command |
Description |
DeepSeek Autocomplete: Toggle |
Enable/disable completions |
DeepSeek Autocomplete: Set API Key |
Store API key in OS keychain |
DeepSeek Autocomplete: Select Model |
Switch model via quick pick |
DeepSeek Autocomplete: Usage Statistics |
View shown/accepted/tokens/cost |
DeepSeek Autocomplete: Debug Log |
Open debug output channel |
Supported Languages
JavaScript, TypeScript, Python, Java, Go, Rust, C, C++, C#, Ruby, PHP, Swift, Kotlin, Scala, Lua, SQL
Each language has custom stop tokens and comment styles for optimal completions.
Privacy & Security
- API key is stored in your OS keychain via VS Code's
SecretStorage — never in settings files or logs.
- Prompts are sent to DeepSeek's API only when generating completions. No telemetry or analytics.
- Sensitive files matching
.env, *secret*, *credential*, *password*, and custom disableInFiles patterns are skipped automatically.
- No data collection — the extension does not collect usage data, send analytics, or phone home (other than API requests you explicitly enable).
Development
git clone https://github.com/andyanamafaza/deepseek-fim-auto-complete.git
cd deepseek-fim-auto-complete
npm install
npm run compile
Press F5 in VS Code to launch an Extension Development Host.
Scripts
| Script |
Description |
npm run compile |
TypeScript compile |
npm run watch |
Watch mode |
npm run lint |
TypeScript + ESLint |
npm test |
Run E2E tests |
npm run vscode:prepublish |
ESBuild production bundle |
Packaging
npx vsce package
code --install-extension deepseek-fim-auto-complete-*.vsix
FAQ
Q: How is this different from GitHub Copilot?
A: It uses DeepSeek's models instead of OpenAI's. DeepSeek is significantly cheaper ($0.28/M output tokens vs Copilot's $X), and the FIM mode is specifically designed for code infilling.
Q: My completions aren't showing up.
A: Check that (1) you've set your API key, (2) the extension is enabled (status bar shows checkmark), and (3) you're in a supported file type.
Q: Can I use my own DeepSeek endpoint?
A: Yes! Set deepseekFim.baseUrl to your custom endpoint URL.
License
MIT
