Tutorial
Set up your first agent
On this page
In this tutorial, you create an agent to answer new employees’ questions and try it in a test channel. You use the Admin console and Slack or Microsoft Teams. No previous experience with Conclude agents is needed.
1. Before you start
You need:
- A Conclude account with access to the Dashboard. Sign in before you start. If you don’t have an account, see Set up Conclude
- Agents – AI teammates in your channels turned on under Admin console > Solutions . An account admin turns it on. See how to turn on solutions
- The Manage agents permission. Everyone has it unless an admin has restricted it. See who can manage agents
- A Slack or Microsoft Teams channel to test in, for example
#agent-test. You must be a member of it
What you should see: Conclude Agents in the Admin console sidebar, with the pages Agents, Deployment, Learning and Knowledge. Account admins also see Usage.
2. Create the agent
- Open Admin console > Conclude Agents > Agents and select Create agent.
- Under Choose a starting point, select Onboarding guide. The text box fills in with a description of that agent.
- Change the description so that it fits your team. You can also fill in Company name and Tone.
- Select Generate. Conclude generates a draft and opens the New agent editor.
- Read the Name, the Role and the Instructions, and change what doesn’t fit. Keep the Model that Conclude chose.
- Select Create.
Nothing is saved until you select Create. See Create an agent for the other starting points.
What you should see: the agent in the list with its role and model. Its Mode is Mention, and its Deployment column is empty.
3. Deploy the agent to a channel
- Open Admin console > Conclude Agents > Deployment and select Deploy agents.
- In step 1, choose your test channel. If the dialog offers App and Channel, select Channel first. If you have several workspaces, select the workspace first.
- Select Next. In step 2, select your agent under Agents.
- Select Deploy.
What you should see: a row for the channel on the Deployment page. It shows your agent, its model and the mode Mention, with a New label.
4. Talk to the agent
In the test channel, write @ and the agent’s name, followed by a question. For an agent called Ada, write @Ada what can you help me with?. Type the whole name. A new agent doesn’t appear in the @mention list of Slack or Teams.
In Slack the agent answers in the thread, or in the channel when your message isn’t in a thread. In Teams it replies under your post. For the next 30 minutes it follows that conversation, so you can ask again without mentioning it.
What you should see: a reply from the agent, under its name and avatar.
If nothing happens, open the channel’s row on the Deployment page and check that the agent’s mode isn’t Off. An account admin can also check under Admin console > Conclude Agents > Usage that the included quota isn’t used up.
5. Give the agent knowledge
So far, the agent relies on its instructions and the model’s existing knowledge. Add a document so it can answer questions about your organization.
- Open Admin console > Conclude Agents > Knowledge and select Add collection.
- Enter a Name, such as
Employee handbook, and a Description. Select Create. - Under Add websites and files, upload a Markdown, text, HTML or DOCX file. You can also enter a web address and select Add website.
- Wait until the line under the collection’s name says that the source is ready.
- Open your agent on the Agents page and go to its Knowledge tab. Tick the collection and select Save.
Then ask the agent a question that the document answers.
What you should see: an answer based on your document. Agents are told to cite the sources they use.
6. Choose when the agent answers
The agent now answers only when someone @mentions it. Change its mode in the channel to let it respond when addressed by name or answer relevant questions without being asked directly.
- On the Deployment page, select the channel’s row. A drawer opens.
- Select the mode in the agent’s row and choose Conversational or Autonomous.
- Select Save.
Modes explains what each mode does.
What you should see: the new mode in the channel’s row. With Conversational, the agent answers “Ada, where do I find the handbook?” without an @.
Next steps
- Deployment explains how to manage agents across channels
- Learning lets a channel with an app learn from its closed tickets and conversations
- Linked accounts makes the agent a channel member that people @mention like a colleague
- Models, keys and usage covers the included quota and your own provider keys