Hermes Agent Completion Hooks Overview | Generated by AI
Here’s the summary.
Yes, there are hooks that fire after Hermes finishes. Three systems:
1. Gateway event hooks (agent:end)
The agent:end event fires when the agent finishes processing a message. Only works in the gateway (Telegram, Discord, etc.), not the CLI. Create a hook directory at ~/.hermes/hooks/<name>/ with HOOK.yaml + handler.py.
# HOOK.yaml
events:
- agent:end
The context dict includes platform, user_id, session_id, message, and response.
Docs: https://hermes-agent.nousresearch.com/docs/user-guide/features/hooks#gateway-event-hooks
2. Plugin hooks (CLI + Gateway)
These work everywhere. The most relevant ones for “agent finished”:
| Hook | When |
|---|---|
post_llm_call |
After the tool-calling loop completes for one turn |
transform_llm_output |
Same timing, but can replace the response text |
on_session_end |
Session is ending |
on_session_finalize |
CLI/gateway tears down session (flush, save, stats) |
Register in a plugin via ctx.register_hook("post_llm_call", my_callback).
3. Shell hooks (CLI + Gateway)
Same events as plugin hooks, but defined in config.yaml and run as shell subprocesses. Writing a script is also an option — JSON piped to stdin, JSON on stdout.
# ~/.hermes/config.yaml
hooks:
post_llm_call:
- command: "~/.hermes/agent-hooks/on-agent-done.sh"
The user doesn’t need to cancel or interrupt. In the CLI, the prompt loop (process_loop in cli.py) naturally returns to waiting for input after every agent response. _agent_running is set to False in a finally block (line 13555 in cli.py), and the loop goes back to _pending_input.get(). The next message you type just starts the next turn.
If you want to be notified when the agent is done so you know it’s your turn to respond, you can enable the terminal bell:
hermes config set display.bell_on_complete true
This plays \a (terminal bell) whenever the agent finishes a response — works over SSH too.