Octop is an open-source, self-hosted AI assistant platform designed for individuals and small teams, emphasizing privacy through a local-first architecture. Built on Python 3.12+ and the FastAPI framework, the system operates via a single-process model that manages a web dashboard, CLI, and various IM integrations (including Feishu, DingTalk, Telegram, and WeCom) using a unified control-plane. The architecture leverages the Harness stack, comprising harness-agent for model routing, harness-gateway for message normalization, harness-memory for hierarchical recall, and harness-browser for CDP-based automation. Users can deploy specialized multi-agent teams with distinct MBTI personas and expert libraries, utilizing RAG for document-based knowledge retrieval. Security is maintained through JWT multi-user isolation, PII redaction, and shell command guardrails. Data is stored locally under ~/.octop/ , defaulting to a SQLite (WAL) database, though PostgreSQL is supported. Octop also features Agent Client Protocol (ACP) support for bidirectional IDE integration with tools like Claude Code, and provides programmable access via HTTP, SSE, and WebSocket APIs. The platform supports diverse backends, including local disk, Docker, and S3, ensuring high portability and flexibility for automated cron tasks and web automation.