A personal, fully local LLM assistant for exploit development, malware analysis, and reverse engineering. It runs entirely on your own machine through Ollama, pairs the model with a small set of agent tools, and keeps a searchable knowledge vault of your own research notes, source, and PDFs.
Nothing leaves the host. No API keys, no cloud calls.
shellcoder.py— interactive agent loop. Talks to the local model, supports both Ollama's native tool-calling and a text-based<tool_call>fallback for models that don't expose tools.tools.py— the agent's tools:read_file,write_file,run_bash(asks for confirmation before running anything), andsearch_vault.data_vault.py— the RAG pipeline. Ingests code, text, PDFs (with OCR fallback for scanned pages), images, and.docxinto a local Chroma vector DB. Watches folders for changes and re-indexes automatically.rag_engine.py— thin query helper the agent uses to hit the vault.Shellcoder.Modelfile— the Ollama model definition. Edit the identity, parameters, and purpose here, then build it (see below).
- Python 3.10+
- Ollama installed and running
- Tesseract and Poppler on your PATH (only needed for PDF/image OCR)
Install the Python deps:
pip install -r requirements.txtBuild the model from the Modelfile (the name must stay Shellcoder unless you also change
MODEL_NAME in shellcoder.py):
ollama create Shellcoder -f Shellcoder.ModelfileStart the agent:
python shellcoder.pyFeed the knowledge vault and query it directly:
# Ingest a single file
python data_vault.py --ingest ./shellcode.c
# Bulk-ingest a folder (recursive)
python data_vault.py --ingest-dir ./research/
# Watch a folder and auto-ingest on save
python data_vault.py --watch ./reverse_engineering/
# Ask a question against everything you've ingested
python data_vault.py --query "What is the SEH offset mentioned in the PDF?"
# Drop chunks for files that no longer exist on disk
python data_vault.py --purgeTune the model's behavior, temperature, context window, and persona in
Shellcoder.Modelfile — the comments in that file explain each knob.
ShellCoder is built for authorized security research, education, and testing on systems you own or have explicit permission to assess. It is dual-use tooling: the same techniques that help you understand and defend systems can cause harm if misused.
You are responsible for how you use it and for complying with all applicable laws and regulations in your jurisdiction. The author accepts no liability for misuse or for any damage resulting from use of this project. Don't point it at anything you aren't allowed to.