Dspilot
English | 简体中文
AI inline code completion for VS Code, powered by DeepSeek or any other OpenAI-compatible model. Requests go straight to the endpoint you configure. GitHub Copilot is not required.
Features
- Inline completion (FIM): context-aware ghost text from the code before and after the cursor, similar files and recent edits, with single-line and multi-line suggestions.
- Fewer wasted requests: debounced requests, cancellation of stale requests, and a local cache that serves repeated or already-typed suggestions without calling the API.
- Next-edit suggestions (optional): predicts your next edit. Requires VS Code's
inlineCompletionsAdditions proposed API.
- Usage statistics: hover the status bar icon (bottom right) to see requests sent, cache hits, requests saved by debounce, and token usage for the current session.
- Safe by default: sensitive files such as
.env, private keys and certificates are always excluded.
Quick Start
- Open Settings and search for
@ext:XiaoHanys.deepseek-autocomplete, or click the status bar icon at the bottom right.
- Set Base Url, Api Key and Model.
- Start typing in any file. Suggestions appear as ghost text; press
Tab to accept.
Example for DeepSeek:
{
"deepseekAutocomplete.completion.baseUrl": "https://api.deepseek.com/beta",
"deepseekAutocomplete.completion.apiKey": "sk-...",
"deepseekAutocomplete.completion.model": "deepseek-chat"
}
Using other models
Any OpenAI-compatible service works: OpenAI, Ollama (http://localhost:11434/v1), vLLM, LM Studio, OpenRouter and so on.
- Native code-completion models (for example Qwen Coder base models): set
apiType to completions and pick the matching fimTemplate.
- Chat models: set
apiType to chat-completions.
- The base URL includes the version prefix (such as
/v1); /completions or /chat/completions is appended automatically.
- Leave
apiKey empty for anonymous local servers. An empty model sends no requests.
Settings
All settings use the prefix deepseekAutocomplete.completion. and take effect without reloading VS Code.
| Setting |
Default |
Description |
enabled |
true |
Turn autocomplete on or off. |
baseUrl |
https://api.openai.com/v1 |
OpenAI-compatible API base URL. |
apiKey |
empty |
API key. Stored in plain text in User Settings; never commit it. |
model |
empty |
Model ID. |
apiType |
completions |
completions or chat-completions. |
fimTemplate |
suffix-param |
FIM prompt format: suffix-param, fim, codegemma, qwen2.5-coder. |
maxTokens |
256 |
Maximum output tokens per request. |
candidateCount |
1 |
Number of candidates when cycling suggestions. |
temperature |
0 |
Sampling temperature. |
timeoutMs |
10000 |
Maximum request duration in milliseconds. |
debounceMs |
150 |
Wait after typing before sending a request; 0 disables it. |
extraHeaders |
{} |
Extra HTTP headers; ${APIKEY} is replaced with your key. |
extraBody |
{} |
Extra fields merged into the request body. |
disabledGlobs |
[] |
File patterns where autocomplete is off. |
enabledLanguages |
{"*": true} |
Per-language switch, for example {"markdown": false}. |
respectSelectedCompletionInfo |
true |
Account for the selected IntelliSense item. |
nes.enabled |
false |
Enable next-edit suggestions. |
nes.model |
empty |
Model for next edits; empty uses model. |
nes.maxTokens |
4096 |
Maximum output tokens for next edits. |
nes.cursorPrediction |
false |
Predict where the cursor should go next. |
nes.eagerness |
auto |
auto, low, medium or high. |
nes.strategy |
copilotNesXtab |
Next-edit prompt format. |
Connection settings (baseUrl, apiKey, timeoutMs, extraHeaders, extraBody) are user-level only, so a workspace cannot redirect requests that carry your key. apiKey and extraHeaders are excluded from Settings Sync.
Status Bar
The icon at the bottom right shows the state: ready, disabled, misconfigured or generating. Hover it for this session's statistics. Token counts appear only when the API reports usage; streamed responses usually omit it unless your server supports stream_options (you can add it through extraBody).
Requirements
VS Code 1.115.0 or later. Regular completion uses stable APIs only; next-edit suggestions need the inlineCompletionsAdditions proposed API and are skipped when it is unavailable.
Development
npm install
npm run package # compile and build a .vsix
