agenticx

AgenticX is a unified, production-oriented multi-agent technology stack. It provides a Python SDK and `agx` CLI for building agents, a Studio server exposing a FastAPI REST API and SSE backend, and Near Desktop, an Electron + React local-first workspace. The project also describes an AgenticX Enterprise path with a web portal, admin console, and Go AI Gateway for authentication, policy, quota, model relay, governance, and audit. Core capabilities include an agent execution engine with a Meta-Agent CEO dispatcher, an embeddable ReActAgent SDK primitive with typed event streaming and parallel tool execution, graph-based orchestration, a unified tool system with MCP Hub and OpenAPI support, hierarchical memory with Mem0 integration, and adapters for 15+ LLM providers. It supports A2A inter-agent communication, MCP resource access, Pydantic-based output validation, and a skill ecosystem with lifecycle management and self-evolution. Knowledge and retrieval features include document processing, GraphRAG, multi-brain knowledge mounting, hybrid vector and BM25 retrieval, and multiple embedding backends. The platform includes long-horizon autonomous coding support with task polling, isolated workspaces, project state machines, and auditable implement-verify-commit loops. Security and operations tooling covers leak detection, sanitization, injection detection, policy evaluation, tool guardrails, approval flows, and sandbox backends such as Docker, Microsandbox, subprocess, and remote HTTP. Observability includes callbacks, Prometheus/OpenTelemetry integration, trajectory analysis, and an evaluation framework with LLM judges and span evaluators. Storage defaults to a `~/.agenticx` file tree with SQLite metadata and optional Redis sessions; Chroma is the default vector backend for the desktop knowledge base. The README includes installation instructions for PyPI and source installs, CLI examples, Python agent and tool usage examples, and links to numerous example projects covering multi-agent collaboration, workflows, A2A communication, memory, human-in-the-loop, GUI automation, intent recognition, ChatBI, deep research, GraphRAG, and more. A security advisory warns about malicious LiteLLM releases 1.82.7 and 1.82.8 and recommends upgrading to a safe version.