PyLume AI
PyLume AI is a very small VS Code extension that provides Python code completions as native inline ghost text. Press Tab to accept a suggestion.
It connects directly from the VS Code extension host to the OpenAI Responses API. It has no login button, browser sign-in, OAuth callback, MFA flow, webview, telemetry, external backend, Docker, or local model dependency.
Install
From a terminal:
code --install-extension pylume-ai.vsix
Or in VS Code:
- Open Extensions.
- Select … in the Extensions view.
- Choose Install from VSIX….
- Select
pylume-ai.vsix.
The target computer does not need Node.js or npm after the VSIX is installed.
One-time setup
- Open the Command Palette with
Ctrl+Shift+P.
- Run PyLume AI: Set API Key.
- Paste your OpenAI API key and press Enter.
- Open a
.py file and start coding.
The key is stored with VS Code SecretStorage and reused automatically. You only need to enter it again if you remove, revoke, or replace it.
An OpenAI API key and API billing/credits are required. A ChatGPT subscription does not automatically include OpenAI API usage.
Commands
PyLume AI: Set API KeyPyLume AI: Remove API KeyPyLume AI: Test ConnectionPyLume AI: EnablePyLume AI: DisablePyLume AI: TogglePyLume AI: Trigger CompletionPyLume AI: Show Status
There are deliberately no login or sign-in commands.
Settings
| Setting |
Default |
Purpose |
pylumeAI.enabled |
true |
Enables Python inline completions. |
pylumeAI.model |
gpt-5.6-luna |
Selects the OpenAI model. |
pylumeAI.debounceMs |
350 |
Waits after typing to reduce requests. |
pylumeAI.maxContextLines |
150 |
Limits lines before the cursor. |
pylumeAI.maxCompletionTokens |
200 |
Limits completion length and cost. |
pylumeAI.temperature |
0.1 |
Keeps results predictable. |
pylumeAI.enableProjectContext |
false |
Optionally includes short excerpts from up to two other open Python files. |
pylumeAI.requestTimeoutMs |
10000 |
Cancels slow requests. |
Privacy and network behavior
- Only Python context near the cursor is sent by default: up to 150 lines before it and 40 lines after it.
- Project-wide indexing is not used.
- Optional project context is off by default and, when enabled, only samples up to two already-open Python files.
- Common secret filenames are excluded and likely inline secrets are redacted.
- The API key is never stored in settings, source code, Git, logs, or the VSIX.
- Runtime requests go directly to
https://api.openai.com/v1/responses.
If the laptop uses an allowlist, permit the VS Code extension host to reach api.openai.com over HTTPS.
Build from source
npm install
npm test
npm run package
License
MIT
