Explain Code Inline
Short grey explanations of your code, shown right in the editor as you work.
- Cursor line: put the cursor on a line and a one-line explanation appears above it.
- Selection: select some code and get a short explanation of what it does.
- Instant on revisits: explanations are cached, and by default the whole open file is explained in the background, so most lines show their explanation without waiting.
Works with Claude through the Anthropic API, or with any OpenAI-compatible server running locally (mlx_lm.server, llama.cpp server, vLLM, SGLang, Ollama).
Getting started
With Claude
- Get an API key from the Anthropic Console.
- Run Explain Code: Set Anthropic API Key from the Command Palette. The key goes into VS Code's secret storage. You can also set the
ANTHROPIC_API_KEY environment variable instead.
- Open a code file and move the cursor.
The default model is claude-haiku-5-5 with low effort, which is fast and cheap. You can change both in settings.
With a local model
- Start any OpenAI-compatible server, for example
mlx_lm.server --model <model> or ollama serve.
- Set
explainCode.provider to openai-compatible.
- Set
explainCode.openai.baseUrl (including /v1) and explainCode.openai.model to match your server.
Where explanations appear
explainCode.lineExplain sets where explanations appear:
| Value |
Behavior |
above (default) |
Grey text above the line or selection (CodeLens). The code shifts down a line while it's shown. |
endOfLine |
Ghost text at the end of the line. Appears instantly and doesn't move the code. |
off |
No explanations. |
Privacy and cost
Your code is sent to the backend you configure. With explainCode.prefetchFile on (the default), the whole open file is sent in 100-line chunks so explanations are ready before you reach each line, and edited lines are sent again. Each file also gets one short summary request, which is used as context.
To send less, turn off explainCode.prefetchFile. Then only the lines you stop on and the code you select are sent, along with a few surrounding lines (explainCode.contextLines). To keep code on your machine, use a local OpenAI-compatible server.
Prose and log files (Markdown, plain text, logs) are never sent.
Settings
| Setting |
Default |
Description |
explainCode.provider |
anthropic |
anthropic or openai-compatible. |
explainCode.anthropic.model |
claude-haiku-5-5 |
Claude model ID. |
explainCode.anthropic.effort |
low |
How much Claude thinks before answering: low to max. |
explainCode.anthropic.fastMode |
false |
Faster output at a higher price, on models that support it. |
explainCode.openai.baseUrl |
http://localhost:8080/v1 |
OpenAI-compatible server URL, including /v1. |
explainCode.openai.model |
default |
Model name sent to that server. |
explainCode.openai.apiKey |
|
Optional bearer token for that server. |
explainCode.lineExplain |
above |
above, endOfLine or off. |
explainCode.lineDelayMs |
150 |
How long the cursor must rest on an unexplained line before a request is made. |
explainCode.lineMinChars |
6 |
Skip short lines such as } or else:. |
explainCode.selectionDelayMs |
250 |
How long a selection must stay unchanged before it's explained. |
explainCode.minChars |
8 |
Ignore shorter selections. |
explainCode.contextLines |
20 |
Surrounding lines sent as context. |
explainCode.prefetchFile |
true |
Explain the whole open file in the background. |
explainCode.prefetchFileMaxLines |
3000 |
Don't prefetch longer files. |
Commands
- Explain Code: Set Anthropic API Key
- Explain Code: Clear Cache: forget all saved explanations. The cache persists across restarts.
License
MIT
