Most team chat keeps people in one app and AI agents in another; Octo Server is the backend that puts both in the same conversation. It powers the OCTO workplace, where Lobsters, the platform's OpenClaw-powered AI agents, sit in channels as first-class participants that read context, run tool calls, and report back alongside their human teammates. Every request flows through one pipeline: token and cookie authentication, organisation-aware role checks and per-channel permissions, business logic that can start or resume an agent session, then fan-out through WuKongIM for real-time delivery. Channels support threaded subspaces, shared files, and a GROUP.md document that stores the group's standing context where the whole team can edit it. The smart summary service reads busy channels and produces structured digests, so catching up means reading one brief instead of hundreds of messages. External bridges connect conversations to GitHub, WeCom, Feishu, and Loop through adapters, and webhooks push group events to your own systems. The official one-command deployment runs the server with the web client, admin console, task service, WuKongIM, MySQL, Redis, and MinIO behind nginx, scaling as one backend that every client speaks to. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 licensed.