sudohopex/kaligpt

Purpose and Entry Points

HackerX/KaliGPT is a Linux command-line AI assistant aimed at ethical-hacking and cybersecurity workflows. Its installed kaligpt command routes user prompts to selectable AI providers, supports interactive model/tool management, can launch provider web chats, and manages provider credentials. The Python package entry point is agents/__main__.py; package initialization is in agents/__init__.py. Installation is orchestrated by install.sh and platform installers under installers/.

CLI Orchestration

agents/__main__.py is the top-level control boundary for command parsing and modes such as Gemini, Ollama, OpenRouter, ChatGPT, LiteLLM, web launch, API-key setup, update, and provider listing. It dispatches into provider implementations rather than exposing a web service. Provider/model configuration and persistent selection logic are centralized under:

• agents/utils/agent_management.py

• agents/utils/agent_configs.py

• agents/utils/api.config.json

Provider Adapter Layer

Each supported backend has a dedicated adapter module, insulating the CLI and shared agent flow from vendor SDK/API differences:

• agents/gemini.py

• agents/chatgpt.py

• agents/ollama.py

• agents/openrouter.py

• agents/litellm_provider.py

Gemini, OpenAI/ChatGPT, OpenRouter, and LiteLLM are network-backed providers requiring credentials; Ollama targets a local runtime but may also access cloud-hosted models. LiteLLM is the broad-provider integration point, advertised as supporting 100+ vendors.

Agent Conversation and Response Flow

A typical invocation flows from CLI mode/prompt selection to provider/model resolution, prompt construction, model execution, optional tool-call handling, and terminal rendering. Shared interaction responsibilities are separated into:

• agents/utils/prompts.py

• agents/utils/parse_n_print_response.py

• agents/utils/openai_tool_adapter.py

The OpenAI tool adapter is an important compatibility boundary: it likely normalizes tool/function-call semantics for provider responses into the application’s tool execution flow. Ollama-specific tool-thinking capability detection is isolated in agents/utils/ollama_tool_think_support_check.py.

Tool Execution and Web Retrieval

The agent can expose tools to models and execute actions locally or through external retrieval services. Tool implementations are grouped in:

• agents/utils/tools/locals.py

• agents/utils/tools/web_request_framework.py

• agents/utils/tools/opensearchapi.py

locals.py represents host-local actions, creating the principal boundary between model output and the user’s machine. Web requests are mediated by a shared framework, while OpenSearchAPI is a distinct external search service integration. agents/utils/openserp_management.py manages the search-service-facing behavior/configuration.

Configuration, Credentials, and User-Facing Launching

Provider API configuration is stored in agents/utils/api.config.json; agent defaults and model choices are handled through the agent configuration/management modules. The separate web-chat launcher uses:

• agents/web_launcher.py

• agents/utils/web_launcher.json

This path opens official provider web interfaces rather than sending requests through the CLI provider adapters.

Runtime and Delivery

The runtime is Python with dependencies declared in requirements/pip-requirements.txt. Delivery is shell-based: install.sh selects or invokes platform-specific scripts for Arch, Debian-family Linux, and Termux:

• installers/installer.arch.sh

• installers/installer.deb.sh

• installers/installer.termux.sh

The repository also contains a focused LiteLLM integration test at tests/test_litellm_provider.py, reflecting the provider abstraction’s external dependency risk.