Curated context packs for debugging, onboarding, reviews, refactors, and AI handoffs. Contexta analyzes the project first, then exports the most useful context for the job.
Portable on Windows, installable on Linux, and runnable from source with Python.
Contexta builds a context pack instead of dumping files blindly. Depending on pack, mode, and task, the output can include:
- A project summary with detected technologies, entry points, likely purpose, and central modules
- A read-this-first path through the repository
- A main execution flow narrative for the most relevant runtime path
- Core files, supporting files, related tests, and changed-file context
- Relationship maps and hotspot/risk notes
- A curated Markdown payload ready to paste into ChatGPT, Claude, Gemini, Copilot, or another coding assistant
Full mode still preserves raw code payloads. The intelligence is added around the payload, not instead of it.
Contexta is designed for the annoying part of AI-assisted coding: deciding what the model actually needs to see.
Use it when you want to:
- explain a project quickly to another developer or model
- review a change set with nearby context
- debug with changed files and likely hotspots already grouped
- onboard into an unfamiliar codebase
- hand off work between AI tools without rebuilding context from scratch
| Feature | Detail |
|---|---|
| GUI + CLI | Desktop workflow for everyday use, plus command-line usage for scripting |
| Context packs | custom, chatgpt, onboarding, pr_review, risk_review, debug, backend, frontend, changes_related |
| Context modes | full, debug, feature, diff, onboarding, refactor |
| Compression modes | full, balanced, focused, signatures, lean |
| Task-aware output | Shapes the export for explanation, bug reports, code review, risk analysis, refactors, tests, dead-code hunting, or AI handoff |
| Project fingerprinting | Detects stack, frameworks, and project type before selecting files |
| Relationship map | Highlights local dependencies and likely related tests |
| Changed Files + Context | Pulls changed files up and expands into nearby relevant code |
| Selection reasons | Explains why each file was included in the payload |
| Read This First + Main Flow | Makes the pack easier for humans and models to navigate |
| Token guidance | Uses tiktoken-backed estimates for tighter compression and safer pack sizing |
| Inline blob protection | Automatically collapses base64/binary literals in all compression modes — no more 15 000-char icon blobs polluting the export |
| Syntax-aware analysis | Uses tree-sitter plus heuristic fallback to extract symbols across multiple languages |
| Multi-language detection | Recognizes Django, Kotlin Android, Spring Boot, Flutter, Rails, React, and more from project structure and import patterns |
| Build pipeline | Uses Nuitka for Windows and PyInstaller plus a Linux install bundle for Unix builds |
onboarding: start here when you need to understand a project fastpr_review: emphasizes review-oriented context and recent changesrisk_review: highlights likely regression hotspots, broad-impact modules, missing coverage, and maintenance weak spotsdebug: pushes suspicious and changed areas upwardbackend/frontend: bias selection toward that side of the appchanges_related: starts from git changes and expands outwardcustom: leaves all fine-tuning to you
full: fastest orientation plus the full selected code payloaddebug: favors hotspots, changed files, and likely failure pathsfeature: biases selection around the focus querydiff: starts from git changes and nearby contextonboarding: richer explanatory structure for first-time readingrefactor: emphasizes central modules and connected files
full: keeps fuller file bodies and prioritizes fidelity (inline blobs still collapsed)balanced: mixes narrative, excerpts, and full payloads for important filesfocused: trims aggressively around the current task/focussignatures: structural overview for quick scanninglean: minimum token usage — metadata and up to 5 signatures per file (~65 tokens/file)
- Download
contexta.exeorcontexta-setup.exe - Run it
- Pick a project folder
- Choose a pack, mode, task, and compression level
- Create the pack and paste the Markdown into your AI tool
Windows SmartScreen can still warn on unsigned open-source executables.
- Download
contexta-linux.tar.gz - Extract it
- Run
./install.shfor a user-local install, or launch the bundledcontextabinary directly
Some Linux environments may require
python3-tkif running from source instead.
git clone https://github.com/pablokaua03/Contexta.git
cd Contexta
python contexta.pyFrom source, install the runtime dependencies first:
python -m pip install -r requirements.txtSome Linux environments may still require a separate tkinter system package such as python3-tk.
python contexta.py /path/to/project
python contexta.py /path/to/project --pack onboarding
python contexta.py /path/to/project --pack risk_review
python contexta.py /path/to/project --mode debug --task bug_report --focus "auth flow"
python contexta.py /path/to/project --pack pr_review --diff --copy
python contexta.py /path/to/project --task ai_handoff --compression balanced --focus "theme"| Flag | Description |
|---|---|
--hidden |
Include hidden folders/files |
--unknown |
Include files with unrecognized extensions |
--diff |
Prefer git diff context |
--staged |
Use staged changes only |
-p / --prompt |
Add a custom instruction or goal |
--focus |
Bias scoring, ordering, excerpts, and related context around a topic |
--mode |
Context selection mode |
--ai |
AI target profile |
--task |
Task profile |
--compression |
Compression strategy |
--pack |
Preset context pack |
-c / --copy |
Copy output to clipboard |
-o / --output |
Custom output path |
--version |
Print version |
- Usually works well: clear task, explicit output format
- Usually avoid: vague goals with no definition of done
- Usually works well: concise but precise instructions, short examples when helpful
- Usually avoid: mixing architecture analysis and implementation without priority
- Usually works well: structured requests, architecture context plus a scoped goal
- Usually avoid: broad prompts with no prioritization
- Usually works well: broader context with explicit priorities, clear formatting instructions
- Usually avoid: assuming long context removes the need for structure
- Usually works well: explicit files, constraints, and expected final state
- Usually avoid: open-ended requests with no target behavior
| Rough size | Heuristic |
|---|---|
< 8k |
Usually manageable for most chat and coding tools |
8k - 32k |
Often comfortable for mainstream model sessions |
32k - 128k |
Better suited to larger-context sessions |
> 128k |
Consider long-context workflows or a tighter export |
# Windows
.\build.bat
# Linux / macOS
chmod +x build.sh && ./build.shImportant:
- Build the Windows executable on Windows and the Linux package on Linux.
- Windows builds use Nuitka and require Visual Studio C++ Build Tools.
- Linux builds create both
dist/contextaanddist/contexta-linux.tar.gz. - On Debian/Ubuntu, install
python3-tkbefore building.
Build outputs:
- Windows:
dist/contexta.exe - Windows installer:
dist/contexta-setup.exewhen Inno Setup is installed - Linux / macOS:
dist/contexta - Linux install bundle:
dist/contexta-linux.tar.gz
If you want the Linux package without setting up Linux locally, run the GitHub Actions workflow .github/workflows/build-linux.yml and download the contexta-linux artifact.
python -m unittest discover tests/Current suite: run python -m unittest discover tests/ to see the latest total.
- Read-only: Contexta does not modify the scanned project
- No runtime telemetry or network requirement in the app itself
- Scan limits prevent runaway exports
- Embedded binary/blob payloads are intentionally suppressed in focused excerpts
- Runtime dependencies are local analysis helpers (
pathspec,charset-normalizer,tiktoken,tree-sitter, andrapidfuzz), not cloud services
See CONTRIBUTING.md
See CHANGELOG.md
