🚀 Send to AI — Task-Aware Code Bundling, Round-Trip
Bundle your errors and the connected code for any AI, paste the answer back, apply it — no API key.
That's the whole loop, inside your editor:
- 🐛 Bundle Errors — grab every file VS Code is flagging + the exact error messages.
- Paste into ChatGPT, Claude, Gemini — whatever you already pay for (or don't).
- 📥 Apply AI Reply — paste the answer back; it writes the edits to your files, reviewable and undoable.
No web tool that's free of an API key does this — because no web tool can see your Problems panel or write to your files. That's the editor-native moat.
✨ What makes it different (v3.8.0)
🐛 Bundle Errors — task-aware by construction
One click bundles only the files VS Code is reporting errors/warnings on, with the exact diagnostics (src/api.ts:42:7 [ERROR] Property 'id' does not exist…) and a ready-to-paste "please fix these" prompt. Web bundlers can't read your Problems panel. This can.
It captures a snapshot of the Problems panel as it is right now. If the errors look stale or wrong, save the file or run "Developer: Reload Window" / "Restart Language Server" so the diagnostics refresh first.
📥 Apply AI Reply — the round trip
Copy the AI's response (code blocks labelled with file paths), run the command, and pick which edits to apply. Every write is:
- ✅ Reviewable — shows up in the editor and in
git diff
- ✅ Undoable — a single
Ctrl+Z
- ✅ Safe — you choose each file; it creates new ones or replaces existing ones
No more hand-copying snippets back block by block.
Tip: copy the AI's whole reply, not the chat's per-block "copy code" button — the file path and the ``` fences must be on the clipboard for the edits to be detected.
🎯 Relevance ranking — keep what matters
When a bundle must be trimmed to fit a token window, files are ranked by relevance to your actual task — filename/path/content overlap with your prompt, plus structural priors that keep entry points, READMEs and config and demote tests, generated code and lockfiles. Dumb bundlers cut the biggest files; this cuts the least relevant ones.
🔗 Dependency-graph awareness — connected, not orphaned
A file imported by a relevant file inherits part of that relevance, so the local dependencies of what you're working on survive the cut alongside it. Follows relative import/require/from edges in JS/TS and Python. You get complete, connected context — not orphaned snippets.
💰 Free + Pro ($9 one-time)
Works with any AI you already use — it's a companion to your existing chat subscription (or a free one), not a replacement for it. No monthly fee, no API key.
|
Copy-pasting by hand |
Web bundler tools |
Send to AI |
| Bundle errors straight from the Problems panel |
❌ |
❌ can't see your editor |
✅ |
| Apply the AI's reply back to your files |
hand-copy each block |
❌ |
✅ one command, undoable |
| Task-aware relevance ranking on trim |
— |
biggest-file-first cuts |
✅ keeps what matters |
| Dependency-graph aware selection |
❌ |
❌ |
✅ |
| Your code leaves your machine |
to your AI only |
uploaded to their server |
to your AI only |
| Price |
free (your time) |
varies |
free · Pro $9 once |
🆓 Free tier
- Single-file send
- Basic project bundling (up to 50 files)
- Standard prompts and ignore patterns
- 🐛 Bundle Errors + 📥 Apply AI Reply
💎 Pro tier ($9 lifetime)
- Unlimited project bundling
- Advanced token estimation & cost calculation
- 4 output formats (Standard, XML, Compact, Minimal)
- Git integration (bundle only changed files)
- Advanced comment stripping & custom ignore patterns
- Visual file-tree generation
- Priority support
🚀 Upgrade — $9 lifetime · No subscription. Use forever.
📦 Quick start
- Install the extension — the SendToAI icon appears in the activity bar.
- Open the panel and use the buttons, or right-click a file/folder → Send to AI.
- Paste into your favourite AI chat — then copy its reply and hit 📥 Apply AI Reply.
Typical loop
🐛 Bundle Errors → paste into Claude/ChatGPT → copy reply → 📥 Apply AI Reply → Ctrl+Z if you don't like it
🎮 Usage
The fastest path — open the SendToAI view and click:
- 🐛 Bundle Errors — bundle everything with a problem + the messages
- 📥 Apply AI Reply — apply the AI's code blocks back to your files
- Bundle project, pick output format, set a token window, estimate cost
Keyboard shortcuts
Ctrl+Alt+A (Cmd+Alt+A): Send current fileCtrl+Alt+B (Cmd+Alt+B): Bundle entire project
Command Palette
Ctrl+Shift+P → type "Send to AI" → pick any command (including Bundle Errors / Apply AI Response).
⚙️ Configuration
File > Preferences > Settings → search "Send to AI":
{
"sendtoai.licenseKey": "XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX",
"sendtoai.autoOpenAI": false
}
Custom ignore patterns
Create a .sendtoaiignore in your project root — same syntax as .gitignore:
# .sendtoaiignore
secrets/**
*.generated.ts
docs/
🤝 Works with every AI
- ChatGPT (GPT-4, GPT-4 Turbo, and newer)
- Claude (Haiku, Sonnet, Opus)
- Google Gemini (Pro, Ultra)
- Local models (Ollama, LM Studio)
Output formats include an XML mode tuned for Claude and structured analysis, plus Compact/Minimal for tight token budgets.
🔒 Privacy & security
- ✅ Zero data collection — your code stays on your machine
- ✅ No telemetry — all bundling happens locally
- ✅ 0 known vulnerabilities — dependencies patched and pinned (
npm audit clean)
- ✅ Open source — audit it yourself
🎯 Perfect for
- 🐛 Fixing a wall of compiler/linter errors in one paste
- 📊 AI code reviews across multiple connected files
- 🔄 Refactoring with full, dependency-aware context
- 🧪 Generating tests from existing code
- 💡 Architecture discussions grounded in the real codebase
📞 Support
📄 License
MIT — see LICENSE.
Bundle your errors + connected code for any AI, paste the answer back, apply it — no API key. Free tier, Pro is a single $9. 🔧
