Overview
hermes chat starts an interactive conversation with Hermes Agent. With no arguments it opens the default REPL/TUI. Pass -q to run a single prompt and exit. You can override the model, provider, toolsets, skills, reasoning effort, session handling, and safety/approval behavior.
Usage
hermes chat [OPTIONS]
hermes # shorthand for hermes chat
Options
| Flag | Description |
|---|---|
-q QUERY, --query QUERY |
Single query (non-interactive). |
--image IMAGE |
Attach a local image to a single query. |
-m MODEL, --model MODEL |
Override the model for this session. |
-t LIST, --toolsets LIST |
Comma-separated toolsets to enable. |
--reasoning LEVEL |
Reasoning effort: none, minimal, low, medium, high, xhigh, max, ultra. |
-s SKILLS, --skills SKILLS |
Preload one or more skills (comma-separated or repeat). |
--provider PROVIDER |
Override the inference provider. |
-v, --verbose |
Verbose output. |
-Q, --quiet |
Quiet mode for programmatic use (suppresses banner, spinner, tool previews). |
--resume SESSION_ID, -r SESSION_ID |
Resume a previous session by ID. |
--no-restore-cwd |
Do not change into the resumed session’s recorded working directory. |
--continue [SESSION_NAME], -c [SESSION_NAME] |
Resume a session by name, or the most recent if omitted. |
--worktree, -w |
Run in an isolated git worktree (for parallel agents). |
--accept-hooks |
Auto-approve unseen shell hooks without prompting. |
--checkpoints |
Enable filesystem checkpoints before destructive file operations. |
--max-turns N |
Maximum tool-calling iterations per turn (default 500 or agent.max_turns). |
--yolo |
Bypass all dangerous command approval prompts. |
--pass-session-id |
Include the session ID in the system prompt. |
--ignore-user-config |
Ignore ~/.hermes/config.yaml and use built-in defaults. |
--ignore-rules |
Skip AGENTS.md, SOUL.md, memory injection, and preloaded skills. |
--safe-mode |
Disable all customizations (config, rules, plugins, MCP servers). |
--source SOURCE |
Session source tag (default: cli). |
--tui |
Launch the modern TUI. |
--cli |
Force the classic prompt_toolkit REPL. |
--dev |
With --tui: run TypeScript sources via tsx. |
Examples
# Interactive session
hermes chat
# Single query
hermes chat -q "Explain the project structure"
# Use a specific model and toolsets
hermes chat -q "Refactor the auth module" -m anthropic/claude-sonnet-4 -t file,terminal
# Attach an image
hermes chat --image ~/screenshot.png -q "Describe this UI"
# Resume the most recent session
hermes chat -c
# Resume a named session
hermes chat -c "debug session"
# Preload skills and run in isolated worktree
hermes chat -s git-workflow,code-review --worktree
# Enable checkpoints and bypass approvals for automation
hermes chat --checkpoints --yolo -q "Run the full test suite"
# Quiet programmatic output
hermes chat -Q -q "Return just the file list"
# Troubleshooting: fully isolated run
hermes chat --safe-mode
Tips
- Use
-Qwhen piping or scripting; the output contains only the final response and session info. - Use
--checkpointswhen running destructive file operations so you can restore with/rollback. - Use
--reasoningto increase or decrease reasoning effort for a single session. --safe-modeis the fastest way to verify whether a problem is caused by your config, rules, or plugins.
See also
hermes, hermes model, hermes sessions, hermes tools, /help in-session.