`city agent chat`

Usage

city agent chat
city agent chat --to my-agent
city agent chat --to my-agent --message "Summarize this repo"
city agent chat --to my-agent --message "Return JSON" --json

Two modes

Interactive mode

Without --message, you enter a persistent terminal conversation:

city agent chat --to my-agent

Best for:

multi-turn interactive exploration
iterative prompt and behavior testing
manual project conversations

One-shot mode

With --message, the CLI sends one turn and exits:

city agent chat --to my-agent --message "Summarize the current repository structure"

Best for:

scripting
one-off analysis
terminal toolchain integration

Streaming behavior

In normal terminal mode, assistant text streams to stdout as it is generated.
--json remains non-streaming so machine-readable output stays stable.
The command still uses the dedicated local-cli-chat-main session by default.

Target resolution

if --to is provided, it uses that agent name directly
if not, and the terminal is interactive, it prompts from currently running agents

If the terminal is non-interactive and you do not provide --to, the CLI does not have enough information to choose a target safely.

Important limitation

The target agent must already be running.

chat will not auto-start the runtime for you.

Common parameters

--to <name>
--message <text>
--json
--host
--port
--token

Useful interactive commands

/exit
/quit
/clear
/help

A common misunderstanding

Do not treat terminal chat as "one globally shared context across every surface."

By default it uses its own dedicated local CLI session, which is excellent for terminal debugging, but it is still its own context boundary.

Continue with:

city agent chat

Table of Contents