hexstrike_augment

HexStrike Augment builds on top of an existing open-source MCP client for Ollama, combining security testing automation, two retrieval-augmented search systems, and a multi-agent workflow layer. It is designed for authorized security research and penetration testing assistance, running entirely with local language models via Ollama rather than cloud APIs. The README is written in Chinese.

The security testing components come from a submodule called hexstrike-ai, which provides a collection of tools for tasks such as vulnerability analysis. The retrieval-augmented search layer runs in two configurations: a lighter one using ChromaDB with Ollama-generated embeddings, and a more complete pipeline that adds BM25 keyword search, vector search, and result re-ranking, with optional Qdrant support for production use.

A multi-agent mode, activated by typing "ma" in the chat interface, uses a three-layer design: a Strategy agent that breaks down a task, a Selector that chooses relevant tools, and an Executor that carries out the steps. The README states this approach reduces token usage by avoiding loading tools that are not relevant to the current task.

Setup requires Python 3.10 or newer and a local Ollama installation with a model that supports tool calling, such as qwen3. Configuration is handled through JSON files copied from provided example templates, with the repository's absolute path filled in by the user. Qdrant and Redis are optional and only needed for the full retrieval pipeline.

The project includes a prominent notice that it is intended for authorized security testing and security research only, and users should not direct its scanning capabilities at systems they do not have permission to test.