Everyday chat
Most of what people do with Trellaris is a conversation. These are the small things that make a conversation worth coming back to: rating an answer, asking about one sentence of it, saving a prompt you keep typing, sharing a thread, choosing how hard the agents should think, and a brief that is ready before you ask.
On every answer
| Control | What it does |
|---|---|
| 👍 / 👎 | Rate the answer. A 👎 offers an optional reason and a comment. Only the person who asked may rate; the rating feeds the agent's analytics and the thumbs_down breaker signal, and the comment is DLP-redacted and scrubbed with the conversation. |
| Quote | Select a passage and ask about just that. The quote travels with your question so the agent answers the sentence, not the whole thread. |
| Follow-ups | Three suggested next questions, generated when you ask for them rather than on every turn, and charged to the answer's own usage row. |
| Run this every… | Turn the turn into a schedule: the same question to the same agent, every weekday at nine, every Monday, every first of the month. A phrase the parser cannot express exactly is declined rather than approximated, and the schedule appears on Automate like any other. |
| Share conversation | A redacted snapshot to the organisation, a team or named people. A recipient may Continue it, which forks into their own conversation so anything further runs as them, with their permissions. Anyone not in the audience gets a 404. |
Saved prompts
A prompt you keep typing becomes a /command. Type / in
the composer to pick one; Settings → Prompts keeps your own
library, and an organisation admin keeps a shared one every member sees.
Official sources
An admin can mark a knowledge source authoritative. Its chunks
rank first in retrieval, the agent sees them tagged [official], and
the citation pill in the answer says Official, so a reader can tell the
policy page from a draft someone uploaded.
How hard to think
| Effort | Behaviour |
|---|---|
| Quick | The Assistant answers directly. No routing, no plan. |
| Standard | The default: the router decides whether one agent can answer or a plan is needed. |
| Deep | Always plans. Above the organisation's cost threshold the plan is shown in the chat for you to approve before anything runs; an admin can set that threshold to zero so every deep plan asks first. |
My usage on Analytics shows what your own conversations cost, by day and by agent.
The brief
The chat console has a Today tab, opened on your first visit of the day. Its live layer needs no model: what needs you (approvals, questions an agent asked), your meetings, your tasks on workspace boards, the follow-ups agents scheduled for themselves, and what finished overnight. An opt-in model layer adds a short written brief built once a day from your mail and calendar, read through your own connections and never through anyone else's.
Meeting prep
Off by default. For each meeting, a note before it starts: who is coming, what you last discussed with them, what is open. External attendees are looked up on the public web and in your own data, never in a colleague's.
Weekly update
Off by default. Once a week a draft of what you did, from your runs, tasks and calendar. It is written for you to edit; Create email draft puts it in your Gmail drafts through the ordinary approval gate, and nothing is ever sent.
Built-in schedules
These run as system schedules on Automate, labelled Built-in: visible, pausable, not counted against your automation allowance, and not editable there. Turn them on under Settings → Chat and Settings → Meetings.