Tutorial

Your first chat

The quick start ran a single turn. In practice you'll use the interactive chat — a multi-turn session that remembers the conversation, can use tools, and is saved so you can pick it back up.

Start a session

Set a provider key, then start agent chat. The four scope flags are the same as everywhere else; here they are written as ...scope to stay readable.

export ANTHROPIC_API_KEY=sk-ant-...

codel00p ...scope agent chat \
  --provider anthropic \
  --model claude-opus-4-8

codel00p prints a banner and waits for you at a prompt:

codel00p chat — provider anthropic model claude-opus-4-8 (session 01JABC...)
Type a message and press Enter. Use /help for commands, /exit to quit.

you> What does this project do, and how is it organized?

Type a message and press Enter. The reply streams back token by token, then you land at another you> prompt. The whole conversation stays in context, so your next message can build on the last answer without repeating yourself.

In-session commands

Anything starting with / is a command, not a message to the model:

/help            Show available commands
/history         Show the current conversation
/tools           List the tools available this turn
/model [id]      Show or switch the model for later turns
/memory          Show approved project memory in context
/session         Show the current session id
/sessions        List all saved conversations
/reset           Start a new conversation
/exit, /quit     Leave the chat

That is the whole loop: talk, let it work, and keep going. The next page gives the agent the tools to actually change your repository.

Note

Every command needs the scope flags (--memory-db, --organization-id, --project-id, --project-name) and a provider key in the environment. See Installation and Quick start if you have not set those up yet.