Create an LLM-ready project context (repo tree + source code) from any project and auto-split it into upload-safe Markdown chunks. Ideal for starting a new ChatGPT chat with your entire codebase available.
- Pure Python (no 3rd-party deps)
- Respects project-level
.gptignore(like.gitignorebut for LLMs) - Optional secret redaction (in-place or into a mirrored "safe copy")
- Output organized under a single folder:
projects_context/<project>_context/and<project>_context.zip
For a project at /path/to/my_project:
projects_context/
└── my_project_context/
├── project_context.md # Repo tree + included file contents
├── manifest.json # Metadata (num files, sizes, chunks)
├── chunks/
│ ├── 0001.md
│ └── 0002.md (only if needed)
└── my_project_context.zip # Everything zipped for single-file upload
If your project is small, you’ll likely get just chunks/0001.md.
- Clone this repo and enter it:
git clone https://github.com/yourname/llm-code-context-generator.git
cd llm-code-context-generator-
(Optional) add a
.gptignoreto your target project (template underexamples/.gptignore). -
Redact secrets (optional):
# In-place (modifies files that match the patterns) python3 src/redact_secrets.py /path/to/your/project # OR: mirror mode (makes a *copy* with redactions; originals untouched) python3 src/redact_secrets.py /path/to/your/project \ --mirror-out /tmp/project_sanitized
- Pack the project:
# Pack originals python3 src/pack.py --root /path/to/your/project --force-include templates/ # OR pack the sanitized copy python3 src/pack.py --root /tmp/project_sanitized --force-include templates/
Outputs will appear under:
projects_context/<project_name>_context/
In a new chat:
- Upload
projects_context/<project>_context/<project>_context.zip- or upload all files in
chunks/(in order: 0001.md, 0002.md, ...)
- or upload all files in
- Paste this starter prompt:
I’ve uploaded my full project context as Markdown (repo tree + code).
Instructions:
- Treat this file as my authoritative project context.
- When referring to files, always cite their relative repo paths (e.g., `apps/accounts/views.py`).
- Do not "summarize away" code unless I ask — reason from the actual code provided.
- If you’re unsure about a detail, ask me which file/chunk to open; don’t hallucinate.
- If something isn’t in the context, say so explicitly.
Confirm you’ve loaded the project context and are ready for questions.
pack.py
--root PATH Project root to scan (required)
--out-parent PATH Parent output dir (default: projects_context)
--project-name NAME Override output naming (defaults to folder name)
--max-bytes N Max bytes per chunk (default ~9MB)
--include-exts CSV File extensions to include (defaults are sensible)
--force-include CSV Path prefixes to always include (e.g., templates/,docs/)
redact_secrets.py
root Project root to scan (required)
--mirror-out PATH If set, write a redacted copy to this directory
Notes
- Chunking is UTF-8 safe (won’t break multibyte chars).
- Huge single files are truncated conservatively inside the pack.
.gptignorecan exclude large/binary/secrets to save tokens.
I only got 0001.md. Is that normal?
Yes—your project fit under the chunk size. One chunk is perfect.
Does ChatGPT remember my repo next time?
No. Start a fresh chat and re-upload the pack when needed.
Can I include images or PDFs?
Usually skip them—they’re large and not token-efficient. If needed, remove them from .gptignore and re-pack.
I don’t want to modify my source during redaction.
Use --mirror-out to redact to a copy and pack the copy.
PRs welcome! Ideas: dry-run listing, richer redact patterns, tests for ignore rules/chunking behaviors.
MIT (see LICENSE).