Skip to main content

Chat

What it is for​

Talk to an agent directly, without waiting for a schedule. Useful for trying a prompt before you automate it, and for asking an agent about something once.

How to reach it​

Chat in the navigation. Pick an agent, start a conversation, type.

The AI companion is the same conversation from anywhere else in Nextcloud. Click the hex button in the corner of any page and you are talking to the same agents, with the page you are on as context.

What the agent can see​

A turn carries what you typed, the conversation so far, and whatever grounding the agent is configured for: attached context objects, RAG over your files and OpenRegister objects, and its own memory.

From the companion it also carries where you are. On a document, the agent knows which document. Without that it would answer questions about "this file" with no idea which file you meant.

What it may do​

The same tool grants that apply to a scheduled run apply here. Chat is not a way around the guardrails: an ungranted destructive tool routes to Approvals from a chat turn exactly as it would from a schedule.

A standing goal​

Some work is not done in one answer. In a session, choose Set a goal (the flag in the session header), write the goal in your own words and say how hermiq knows it is reached:

  • A count of objects reaches a target. For example: the overdue permit applications without a reminder, target 0. The count is read as you, so it only sees what you may see. Prefer this one: it needs no model to decide.
  • A model answers a question about the last answer. Use it when there is nothing to count.

Then pick how often the agent takes a turn (every 15 minutes up to once a day) and the most turns it may take (1 to 50). The agent keeps working in the same session, so it sees its earlier turns. Every turn passes the same checks as a scheduled run: when the agent is switched off, the kill switch is on or the budget is used up, the turn waits and the header says the goal is blocked. The header shows the goal, the turn and the last check. You get a notification when the goal is reached or the turns run out. Stop goal ends it; only you and the agent's owner can stop it.

When there is no model​

If no LLM is reachable, the turn fails and says so. It does not hang, and it does not invent an answer.

API​

  • GET|POST /apps/hermiq/api/conversations
  • POST /apps/hermiq/api/chat/message
  • GET /apps/hermiq/api/chat/health
  • GET|POST /apps/hermiq/api/sessions/{uuid}/goal and POST /apps/hermiq/api/goals/{id}/stop

Where to go next​

Happy with how it answers? Give it a schedule from its agent page.