LLM4CVE

LLM4CVE is a command line tool that use LLM to analyze and summarize Common Vulnerabilities and Exposures (CVE). Given a CVE (or a list) the system generate a summary of their description.

Description

The database containing all CVEs is downloaded locally. At every analysis request the database is queried and the JSON corresponding to the CVE is given to an LLM model for summarization (default llama3.2). The results are then given to a bigger model (default deepseek-r1:14b) for complete analysis and printed in markdown format to screen. A flag can be used to specify a file for output save.

Set up

// go run main.go -h
`  _ _           _  _
  | | |_ __ ___ | || |   _____   _____
  | | | '_ ' _ \| || |_ / __\ \ / / _ \
  | | | | | | | |__   _| (__ \ V /  __/
  |_|_|_| |_| |_|  |_|  \___| \_/ \___|`
llm4cve is a CLI tool that analyzes and summarizes CVEs using local LLMs.

Usage:
llm4cve [CVE_ID] [flags]

Flags:
-h, --help                help for llm4cve
-m, --model string        Chose LLM model for analysis ['llama3.2','deepseek-r1:14b'] (default "deepseek-r1:14b")
-u, --ollama-url string   Use custom URL for Ollama API (default "http://127.0.0.1:11434")
-o, --output string       Save output to file. Output will be in Markdown
-v, --verbose             Display additional information

Example of usage :

# Run
go run main.go CVE-2025-29927 CVE-1999-1000 -v -o -u 'http://your-ollama-url:11434'
# Output produced is in output.md

# Build
go build
./llm4cve CVE-2025-29927 CVE-1999-1000 -v

Tools and Technologies

• Langchain Framework
• Ollama (DeepSeek, llama 3.2)
• Go

Future Work

• Custom URL for Ollama as flag
• Custom output file as flag
• Model choice as flag
• Add Support for CWE, Capec
• Add different Database Source

Authors

@GiovanniMenon
@NicoloPellegrinelli