Octop setup: documents, expert teams and OpenCode

The tutorial introduces Octop, an open source, MIT-licensed AI assistant from Tencent Cloud, and builds a release-preparation workflow through its web panel. The presenter covers terminal installation, administrator setup and model connection. Desktop releases are also described for Windows, Mac and Linux. The installer handles Python setup; the web panel uses port 8088.

Octop is self-hosted, but model requests can still leave the machine. The presenter explains that cloud APIs may cost money and that Ollama can supply a local LLM. Hardware needs depend on the model, which must handle tool calls reliably for this workflow. Local ONNX embeddings handle document search separately from the chat model.

A release assistant searches fictional Taskboard notes and cites its sources. The presenter checks those citations against the file, then adds a reviewer to catch invented synchronization support and a release date. Team features are described as beta and require a coordinator plus at least two active members. Stopping the coordinator may leave delegated work running.

The OpenCode example uses ACP to create an announcement file. OpenCode needs its own configuration on the same machine. A weekday 9 AM reminder demonstrates fixed Text tasks versus Agent jobs that invoke a model. Scheduling requires the Octop process to remain running and the correct timezone; the stated default is Asia/Shanghai.