CodeBlendCodeBlend is a Visual Studio Code extension designed to show how much of your code is generated with AI assistance versus manual input, helping organizations understand the overall contribution of AI to team productivity. Think of it as a “fitness tracker” for your coding workflow—it monitors activity in real time and provides insights into AI usage patterns. Organizations can easily access aggregated statistics on team coding workflows without exposing individual employee details.
Key Features
InstallManual VSIX Install
Line Highlight Styles (
|
| Command | Purpose |
|---|---|
| Sign In / Switch Account | Acquire / switch Microsoft account & verify ingest rights |
| Reset Current Session Stats | Start fresh session; snapshot previous |
| Manual Submit If Commit Changed | Force boundary if commit advanced |
| Toggle Character-Level Highlight Style | Toggle between previous line style and char merged spans |
Configuration (Highlights)
| Setting | Default | Summary |
|---|---|---|
codeblend.lineHighlights.style |
border |
Visualization style (border/gutter/mixed/char/none) |
codeblend.compatibilityMode |
false |
Disables the type-command interception fallback to avoid conflicts with other extensions; expect slightly less accurate typing attribution. |
codeblend.maxTrackedDocChars |
200KB |
Max per-document size (in KB) to attribute; files above are ignored (range 1–1024KB). Legacy byte values auto-mapped. |
codeblend.maxCheckpoints |
100 |
Max number of recent attribution checkpoints retained for fast restore (minimum 50, no explicit upper cap). Older ones dropped FIFO. |
Marker semantics: bright = fully AI line, light = mixed line, none = no AI characters. In char mode, a subtle background highlights contiguous AI character spans (merged). Safeguards:
- Spans merged (never per char) & capped (~5k) else fallback to line mode.
- Dense high‑AI very large files auto fallback.
- Debounced updates share line highlight timing.
Data Upload & Privacy
This plugin only records repository and commit metadata—no personal data is collected. Microsoft login is needed to upload data. For details on collected data, visit HowItWorks.
The limited data described below is uploaded solely to understand overall AI assistance usage patterns. It is not intended or suitable for individual performance evaluation, and no personal identity or prompt content is collected.
Commit Summary Schema
Row sent per commit boundary (after sign‑in):
| Column | Description |
|---|---|
reponame |
Repository name (remote origin base or local folder) |
branch |
branch name (Hash Value) |
commit |
HEAD hash |
files |
Array of repo‑relative file paths (prefixed with repo name and hash file name) |
pluginMeta |
Nested plugin metadata structure |
totalCharacters |
Total inserted characters this session/commit |
likelyAiCharacters |
Inline / high confidence AI characters |
pastepossiblyAiCharacters |
Medium confidence AI paste characters |
pastelikelyAiCharacters |
High confidence AI paste characters |
pasteFromCopilotCharacters |
Pasted characters detected as Copilot chat / Ask output |
pastelikelyHumanCharacters |
Pasted characters judged likely human |
typeCharacters |
Manual typing characters |
idecompleteCharacters |
IDE (non‑AI) completion characters |
deletedCharacters |
Characters deleted |
ignoredCharacters |
Explicitly excluded characters (outside tracked scope / suppression) |
fallbackCharacters |
Generic AI bucket when specific heuristic classification fails |
pluginAiCharacters |
Characters attributed via external plugin session marked as AI |
pluginNonAiCharacters |
Characters attributed via external plugin session marked as non-AI |
License
MIT – see LICENSE.
EULA - see EULA.
FAQ
How is AI % computed? (likelyAi + pasteFromCopilot + pastelikelyAi + pastepossiblyAi + fallback + pluginAiCharacters) / totalCharacters. ignoredCharacters are excluded from both numerator and denominator.
Where are all the heuristics & edge cases documented? In the full Deep Dive (counters, suppression window, checkpoints, paste tiers, attribution model, privacy details).