Tencent Cloud open source Octop 1.0.2b4: a multi-user, multi-agent self-hosted AI assistant installed in one process腾讯云开源 Octop 1.0.2b4：一个进程装下多用户、多智能体的自托管 AI 助手

The vast majority of personal AI assistants are "one person, one window": if you want to share them at home or in a small team, you can only use the same account, and the memories are intertwined with each other and the permissions are inseparable.

Octop has a single-administrator multi-user architecture: each member has his own expert team, workspace and portable memory, and the administrator manages models, permissions and channels in a unified manner. The entire set of things runs on your own machine, so there is no need to take it out to change functions for privacy.

Expert database and expert market: Cut experts according to scenarios, and release adjusted experts, skill pools, and sub-agents to colleagues for reuse; 16 built-in MBTI personality templates stabilize the expression style of each agent.

Built-in security: JWT multi-user isolation, tool call approval, shell command guardrail, PII desensitization; workspace backend is pluggable (local disk, Docker sandbox, PostgreSQL, COS/S3) and separated from the control plane database.

The knowledge base supports RAGing local documents; the memory is driven by Octop Memory and migrates with the workspace; the plug-in system can load third-party extensions, and built-in plug-ins can be opened on demand.

Open an interactive terminal directly in the console (AI assists in command execution and troubleshooting); Browser AI+ provides headless Chromium sessions for web page automation, screenshots, and remote browsing; remote desktop supports Linux, Windows, and macOS, and the desktop can be isolated with one click on headless Linux.

The official also provides native Windows / macOS / Linux desktop clients and FnOS installation packages (desktop package size is 160–220 MB, this site does not include single file size control, you can download it on the Releases page).

The development scenario can use the ACP two-way channel: octop acp connects OpenCode / Claude Code and delegates coding tasks with permission confirmation.

The download button provides the 15.7 MB source code package of the official Release v1.0.2b4, which is licensed by MIT and can be freely used in personal and commercial environments; this machine requires Python 3.12 and above.

Startup method: After installing the dependencies, execute octop run, then open the console in the browser and follow the wizard to create an administrator and configure the model. You can also directly pip install octop from PyPI.

This article is compiled from public sources and ships with the matching resource. Product features and pricing are subject to the official page. Resources are for learning and exchange only — please respect the original license.