octop agent
Create, inspect, and control agent lifecycles. Most subcommands work without requiring octop run to be running.
Subcommands
Run
octop agent use <id> to write the agent ID into ~/.octop/cli_state.json. All subsequent commands that accept --agent will default to this value, saving you from repeating it.
octop chats
Manage conversation threads and send messages. The repl and send subcommands start a temporary server process automatically — no separate octop run is needed.
Subcommands
octop chats repl and octop chats send start a temporary in-process server for the duration of the command. You do not need to run octop run first. The list, get, create, update, and delete subcommands work fully offline against ~/.octop/octop.db.octop skills
Enable or disable per-agent skills. All subcommands query the skill library at boot and require a running server.
Subcommands
octop models
Manage the model catalog, configure providers interactively, and pull Ollama models.
Subcommands
octop provider
Manage LLM provider records in the local database. The test subcommand requires a running server to make a live API call.
octop provider list, create, and delete operate offline against ~/.octop/octop.db. Only octop provider test requires a running server, because it sends a live ping to the provider API.octop acp
Expose an Octop agent as a stdio JSON-RPC ACP server. Boots a standalone OctopServer (reads ~/.octop/); does not require octop run to be running.
Options
Point your IDE at it
Configureoctop acp as the ACP server command in your IDE. Zed example:
This is the reverse direction of the ACP runners guide: here your IDE drives an Octop agent, rather than an Octop agent delegating to a CLI tool.

