Chat With an Agent

The simplest path

Make sure the agent is running, then run:

city agent chat

If you do not explicitly specify a target and your terminal is interactive, the CLI will let you choose from running agents.

That default experience is optimized for humans, not scripts.

Target a specific Agent

city agent chat --to my-agent

If you manage multiple agents, using --to explicitly is the most stable and reproducible pattern.

Send one message and exit

city agent chat --to my-agent --message "Summarize this repository"

This is usually the best shape for scripting, one-off analysis, or command-line tooling.

JSON output

city agent chat --to my-agent --message "List today's changes" --json

Use this if another script or pipeline will consume the result.

Interactive session behavior

• it uses a dedicated local CLI main session by default
• every line you enter continues the same terminal session
• it does not automatically share context with every other UI entry point

That last point matters a lot. Terminal chat is not "one universal global conversation."

Common exit commands

• /exit
• /quit

If you get no reply

Check these first:

1. city agent status
2. model binding
3. whether the target agent is actually running

If the error is "daemon unreachable" or something similar, treat it as a runtime-state problem first, not a prompt problem.

Next:

• Connect chat channels
• CLI chat reference