Every AI session starts the same way: hunt down the relevant files, manually skip `node_modules`, build artifacts, and lock files, zip them, find the zip, upload it. Then do it all again next session.
contextzip eliminates that entirely. Run it from your project root — it detects your stack, applies smart exclusions, produces a lean ZIP, and opens your file manager with the archive already selected. One `Ctrl+C` and you're done.
- **Smart framework detection** — automatically identifies Node.js, Next.js, Python, Django, FastAPI, Rust, Go, and Ruby, applying the right exclusion rules for each
- **Respects `.gitignore`** — your existing ignore patterns are honoured automatically
- **Git-aware packaging** — use `--git-changes` to package only modified, staged, and untracked files; perfect for incremental debugging and PR review sessions
- **AI-powered file selection** — describe your task in plain English with `--prompt` and Gemini selects the minimum relevant files automatically, no manual hunting required
- **Terminal error watcher** — wrap any dev server with `contextzip watch` to auto-detect errors and package a ready-to-upload debug context in one keypress
| `get_git_changes(path?)` | Modified, added, and untracked files from git |
| `get_files(path?, include?, exclude?)` | All project files after exclusion rules |
| `create_zip(collection, output?)` | Write a `FileCollection` to a ZIP archive |
| `detect_ecosystem(path?)` | Detect framework and confidence level |
All functions default `path` to `Path.cwd()`. Errors raise typed exceptions (`NotARepositoryError`, `GitNotFoundError`, `NoFilesError`, etc.) rather than exiting.
The `--prompt` flag lets you describe a task in plain English. contextzip scans your project, builds a lightweight file map, and asks Gemini to return the minimum set of files needed for that task — typically 2–5, never more than 10. The result is a tightly scoped ZIP with only what you'd actually open to make the change.
The ZIP also includes a `prompt.txt` describing the task, so when you drop it into Claude, ChatGPT, or any other AI tool, it immediately understands what you're trying to do.
**First-time setup:**`--prompt` requires a free Gemini API key from [Google AI Studio](https://aistudio.google.com/apikey) — no credit card needed. On first use, contextzip guides you through obtaining and saving one. You can also skip the setup entirely with an environment variable:
The `watch` command wraps your dev server, buffers its output, and packages a debug-ready ZIP the moment you spot an error — no manual file hunting, no copy-pasting stack traces.
```bash
contextzip watch -- npm run dev
contextzip watch -- python manage.py runserver
```
contextzip starts your process normally. You see output exactly as you would without it. In the background, it watches the stream for errors. When one is detected, a prompt appears directly beneath the error output:
Press **D** and contextzip immediately writes `.contextzip/debug-context.zip`. Your server keeps running — no restart, no interruption.
**What's in the ZIP:**
| File | Contents |
|---|---|
| `prompt.txt` | Auto-generated: detected framework, error type, and task description — ready to paste into any AI tool |
| `terminal-error.txt` | The cleaned, noise-stripped error block and stack trace |
| `source-files.zip` | Source files referenced in the stack trace, paths preserved |
**On Ctrl+C:** If no errors were packaged during the session, contextzip offers one final prompt to capture the full session output — useful when something looked wrong but didn't match a known error pattern.
**Supported frameworks:** Python, Django, FastAPI, Node.js, Next.js, React. Each has its own error detection patterns and noise filters so the output stays clean across stacks.
> **Note:** `watch` works best with dev servers that don't read stdin interactively (`npm run dev`, `manage.py runserver`, etc.). PTY emulation is not used — on Windows, color passthrough may be limited.
See [CONTRIBUTING.md](https://github.com/akadeepesh/contextzip/blob/main/CONTRIBUTING.md) for local setup, how to add a new framework, and PR guidelines. Please open an issue before starting a large PR so we can align on the approach first.