Customise an agent
An agent is a set of records in the Setup Portal. You can reshape Lumen, or create as many agents of your own as you need, without changing code.
What makes up an agent
Section titled “What makes up an agent”- The Agent record: its name, look, model and settings.
- One or more Topics: each a bundle of instructions and tools for one area of work.
- Instructions: the guidance the agent is given, written in plain language.
- Tools: what the agent is able to do.
See How it’s organised for how these connect.
Create an agent
Section titled “Create an agent”- In the Setup Portal, open Agents and select New.
- Give it a Name. This is the name users see, and the name you use to place it in an app.
- Choose its LLM and enter Max Tokens, for example 10,000. Both are required.
- Leave Active ticked. An inactive agent cannot be used, which is a way to switch one off without deleting it.
- Add at least one topic, as described below.
Give it a look
Section titled “Give it a look”Tick Enable Chat Interface on the Agent record to show the branding fields.
| Field | What it does |
|---|---|
| Tagline | A short line under the agent’s name on the welcome screen |
| Primary Color | The agent’s accent colour, used for the user’s messages, the send button, the agent’s name and the letter avatar. Enter a hex code such as #2A2119. Left blank, it is Salesforce blue |
| Avatar Static Resource | The API name of a static resource holding the agent’s image (PNG or SVG). Left blank, the agent gets a letter avatar |
Refresh the browser page to see a change to these fields.
Tell it what to do
Section titled “Tell it what to do”Description
Section titled “Description”The Description on the Agent record opens the agent’s prompt: “You are Name. Description”. Put the agent’s role and its overall rules there. Lumen’s is a good example to copy from.
Topics
Section titled “Topics”A topic groups the instructions and tools for one job, such as answering questions about records. Create a Topic record with a Topic API Name and a Description of what it covers, then link it to the agent with a Topic Map record.
An agent can have several topics, and a topic can be shared by several agents. The Order on each Topic Map sets the order the topics appear in the agent’s prompt. Untick Active on a Topic Map to take a topic away from an agent without deleting anything.
Instructions
Section titled “Instructions”Each Instruction record is one block of guidance in plain language: how to answer, what to check first, what never to do. Every active instruction in the agent’s topics is sent to the model as part of its prompt.
- Order sets the sequence within a topic. Number them 10, 20, 30 so you can insert one later.
- Untick Active to try the agent without an instruction, then tick it again.
Changes take effect on the next message. There is nothing to deploy.
Link a tool to a topic with a Topic Tool Map record. The agent can use only the tools on its topics, so leaving a tool off is how you stop an agent doing something. See Built-in tools and Add your own tool.
Tune how it behaves
Section titled “Tune how it behaves”These are on the Agent record. The defaults suit most agents.
Several are measured in tokens, the units AI models count text in. A token is a word or part of a word; 1,000 tokens is roughly 750 words. Providers charge by the token, for what the agent sends and for what the model writes back.
| Field | What it does | If left blank |
|---|---|---|
| LLM | The model the agent uses. See Use a different model | Required |
| Reasoning Effort | How much the model thinks before it answers: low, medium, high or disabled. Higher is more accurate, slower and costs more. Use disabled for a model that does not support reasoning | The model’s own default |
| Max Tokens | The longest reply the model may write. Lower it to cap cost | Required. Lumen is set to 10,000 |
| Conversation Depth | How many earlier exchanges in the conversation are sent with each new message. More gives the agent a longer memory and uses more tokens. Set it to 0 for an agent that treats every message as new | 15 |
| Tool Log Depth | For how many recent messages the agent also remembers the tool calls it made and their results. More helps with long multi-step work and uses more tokens | 5 |
| Show Execution Message | Shows the user a short progress line for each tool as it runs, such as “Searching for Northwind Traders account” | Off |
Tool Log Depth cannot be higher than Conversation Depth, and if you set Tool Log Depth you must set Conversation Depth too.
Put it in front of users
Section titled “Put it in front of users”Lumen comes ready in the ApexAgent Setup Portal’s utility bar. For an agent of your own:
- In Setup, open App Manager and edit the Lightning app your users work in.
- Under Utility Items, add the apexAgentChat component. This is the chat window itself; lumenAgent is the same window already set to Lumen.
- Set its Agent Name to the Name of your Agent record.
Users need the ApexAgent User permission set, the same as for Lumen.