talk

The talk command starts an interactive, multi-turn conversation session with an AI agent. Unlike the single-shot ask command, talk maintains context across messages so you can ask follow-up questions naturally without repeating background information.

Usage

clanker talk [flags]

Flags

Flag	Type	Default	Description
--agent	string	hermes	AI agent to use for the conversation

How It Works

When you run clanker talk, an interactive REPL session starts where you type messages and receive responses from the selected AI agent. The session preserves the full conversation history, so the agent understands the context of earlier exchanges when responding to new messages.

The conversation flow looks like this:

$ clanker talk
Hermes Agent (interactive mode)
Type 'exit' or 'quit' to end the session.

you> What EC2 instances are running in us-east-1?
hermes> [response with instance details]

you> Which of those are using the most CPU?
hermes> [response referencing the previously listed instances]

you> Show me the CloudWatch metrics for the top one
hermes> [response with metrics for the specific instance discussed]

Each follow-up question builds on the prior context, which means you do not need to re-specify regions, resource names, or other details already established in the conversation.

Ending a Session

You can end a talk session in any of these ways:

• Type exit or quit
• Type /exit or /quit
• Press Ctrl+D (sends EOF)
• Press Ctrl+C to interrupt the current response (pressing it again will exit)

When to Use talk vs ask

Use talk when...	Use ask when...
You need to explore a topic iteratively	You have a single, self-contained question
Follow-up questions depend on previous answers	You want to use the output in a script or pipeline
You are investigating an incident or debugging	You need a one-shot answer with no follow-up
You want a conversational, exploratory workflow	You want to combine with flags like --maker or --apply

Agents

The --agent flag selects which AI agent handles the conversation. See Agents for a complete overview.

Hermes

The default agent. Hermes is a general-purpose infrastructure agent that can answer questions about your cloud environment, execute tool calls, and maintain conversational context across turns.

# Explicitly select the hermes agent (same as the default)
clanker talk --agent hermes

TIP

The Hermes agent requires an initial setup step. If you see an error about the agent not being found, run make setup-hermes in the clanker repository to install it.

Claude Code

Anthropic's Claude Code CLI with full tool use, file reading/writing, and code analysis capabilities. Requires the claude binary to be installed.

clanker talk --agent claude-code

The session looks like this:

$ clanker talk --agent claude-code
Claude Code Agent (interactive mode)
Type 'exit' or 'quit' to end the session.

you> What EC2 instances are running?
claude-code> Here are your EC2 instances:
| Instance ID | Type | State |
|---|---|---|
| i-0ccdce88a8035bb75 | t3.xlarge | running |

you> Tell me more about that instance
claude-code> [responds with details, using conversation context]

TIP

Install Claude Code with npm install -g @anthropic-ai/claude-code. See the Claude Code agent docs for full setup and configuration details.

Debugging

Enable debug output to see internal agent activity such as tool calls and reasoning steps:

clanker talk --debug

With debug mode enabled, you will see additional output prefixed with [tool: ...] and [thinking: ...] alongside the normal conversation.

Examples

Start a basic conversation:

clanker talk

Start a conversation with debug output:

clanker talk --debug

Specify the agent explicitly:

clanker talk --agent hermes

Use Claude Code agent:

clanker talk --agent claude-code

See Also

• ask -- Single-shot natural language queries
• Agents -- All available agents and setup guides
• Configuration -- Setting up AI providers and credentials