One agent per job. Shared by the whole team.
An agent in tting is a teammate with a job description: what it may read, how it should work, what it may do, and which model it thinks with. Set it up once and everyone in the workspace works with the same one.
What an agent is made of
Every setting is something you would tell a new hire on their first day.
- Name and description. So the team knows what to ask it.
- Instructions. The role, the tone, the boundaries.
- Model. Claude, GPT, Qwen or Kimi. Choose one per agent, and change it for a single conversation when you need to.
- Skills, up to 50. The procedures and output formats it follows. More on skills.
- Knowledge. Which of your documents it may answer from.
- Connectors. Which connected services it may use, always with the permissions of the person asking.
- When it asks for approval. Risky actions only (the default), Always ask, or Never ask.
- Web search. Every agent searches the web automatically when it helps, with no per-agent switch. Web sources get the same numbered citations as your documents.
Versions, so nothing breaks mid-conversation
Editing an agent creates a new version. A conversation already under way finishes on the version it started with. If a change turns out to be a mistake, roll back to an earlier version; editing from there starts a new one.
- Copy any agent. The copy is yours to change.
- Deleting has guardrails. The representative agent and the agents tting provides can't be deleted. Anything else can, by its creator or an admin, and you see what it leaves behind before you confirm.
One representative agent per workspace
Every workspace has one agent that new conversations and new members start with. tting ships one that explains the product and takes you to the right screen. Admins can promote any agent they built to take its place.
- tting's own agents are in every workspace and maintained by tting. They can't be edited or copied. To customize, build your own and make it the representative.
- Switch mid-conversation. Change to another agent from the conversation header. The thread continues.
When it needs a decision, it asks with options
Instead of a wall of text, an agent that needs your call shows a short list. Tap one, or type your own answer, and the conversation carries on from there.
- Bring in another agent with @. It answers that one turn. The conversation stays with the agent you started with.
- If an answer fails, retry it or ask a different model, without starting over.
The conversation is where the work gets done.
Citations you can open
When an answer comes from your documents, each claim carries a number. Tap it to see the original passage.
Files in the conversation
Attach PDFs, documents or images and the agent uses them for that conversation.
Model per conversation
The agent has a default model. Change it for one conversation without touching the agent.
Stop any time, pick up anywhere
Cut a long answer short, or continue the conversation on another device, even while the answer is still being written.
Your conversations are yours
Only the person who started a conversation can see it. Copy an answer, or export the conversation as text or a file, when you decide to.
Grounded answers
On facts about your company, an agent answers only when it has evidence and shows where it came from. If there is none, or sources disagree, it tells you.
Set up your first agent in a few minutes.
Or install one from the marketplace and see how it's put together.