Configuration reference

Understand every major area on the agent Configure page and when to use each option.

Use this page when you want to understand the full Configure surface for an agent. The goal is not to turn every setting on. The goal is to make deliberate choices so the agent is useful, predictable, and safe for the business process it supports.

Open an agent from Agents, then click Configure.

The Configure map has Core settings in the centre and capability areas around it.

Configuration map

Area	What it controls	Use it when
Core	Profile, instructions, AI settings, and dynamic discovery	Every agent
Objects	Business object types and specific object records the agent can access	The agent should read or update structured records
Tools	App actions, built-in actions, workflow actions, and MCP integrations	The agent should do something, not only answer
Knowledge	Knowledge notebooks the agent can search	The agent should answer from documents
Skills	Reusable skill packs, when enabled for your workspace	The agent needs packaged expertise or executable capabilities

Core: Profile & instructions

Open Profile & instructions first.

Profile and instructions decide how people recognise the agent and how the agent should behave.

Option	What it does	Citizen developer guidance
Avatar	Visual identity for the agent	Useful for shared agents, but not required
Name	Label shown in lists, chat, and agent switchers	Use a job-based name, not a person's name
Description	Short explanation of what the agent does	Write who should use it and what business process it supports
Instructions	Behaviour rules sent to the model	Include job, boundaries, escalation rules, source rules, and tone

Good instructions are specific:

text

You help the Finance team review supplier invoice questions.
Ask for missing supplier, amount, or invoice number before taking action.
Do not approve payments. If approval is needed, ask the user to start the approval workflow.
Use the Finance policy notebook for policy questions and say when the notebook does not answer.

Core: AI settings

Open AI settings to choose the model connection and tune runtime behaviour.

AI settings choose the provider/model and control what built-in and platform tools are available.

Option	What it does	Default habit
AI Connection	Approved provider account the agent uses	Choose the admin-approved connection for the team
Model	Specific model used for replies and reasoning	Use the recommended model unless you know why to change
Temperature	Controls randomness; lower is more focused, higher is more creative	Keep lower for policy, finance, and operations agents
Max Tool Roundtrips	Limits how many tool-use loops the agent can perform for one request	Keep modest; raise only for multi-step research or actions
Frequency Penalty	Reduces repetition of the same phrases	Usually leave at default
Presence Penalty	Encourages the model to introduce new topics	Usually leave at default for business agents
Max Retries	Number of retries when the AI call fails	Use a small number for important shared agents
Message Lookback Limit	How many recent chat messages are sent back as context	Raise for long discovery chats; lower to reduce token usage
Task Naming Instructions	Instructions for naming new conversations/tasks	Use when conversation names should follow a team convention

Core: Dynamic discovery

Dynamic discovery controls whether the agent can help maintain its own build setup while you work.

Dynamic discovery is useful while building, but should be reviewed before sharing a stable agent.

Option	What it does	Use carefully when
Agent-maintained instructions	Master mode that lets the agent maintain instruction sections during build work	You are still shaping the agent through conversation
Auto-discover workspace actions	Lets the build kit discover workspace actions	The agent needs to find relevant app/workflow actions
Auto-discover MCP servers	Lets the build kit discover MCP servers	The agent needs to find MCP tools
Instruction progress	Shows whether important instruction sections have content	You want a readiness check for role, task structure, project structure, and rules

For a stable shared agent, review dynamic settings before sharing. If public sharing is blocked, the panel shows the reason.

Dynamic discovery turns on agent builder platform tools (search/propose actions, MCP, skills, knowledge, instructions, and related helpers). It does not give the agent its own triggers. Those belong on workflows. See Builder tools.

If Other still lists Search Available Triggers or Propose Agent Trigger, ignore them. They are leftovers and are not used by the agent builder.

Objects

Open Objects when the agent should use structured business records, not just free-text documents.

Option	What it does
Search	Finds object types or objects
Object type switch	Enables tools for a whole type, such as customers, suppliers, or assets
Pinned object switch	Gives the agent access to a specific record

Use object types for repeatable records. Pin individual objects when the agent should work with a small known set.

Tools

Open Tools when the agent needs to take action.

Tools add actions the agent can call. Add only what the agent needs for its job.

Tools can include:

Built-in Automate actions
App actions from connected services
Workflow actions
MCP integrations

Each tool increases what the agent can do. Add tools gradually and test with prompts that should and should not use the tool.

MCP integrations

MCP integrations are configured from Tools, but require a separate account connection first.

State	Meaning
Not connected	The user still needs to connect the MCP account
Connected	The account is connected on the MCP page
Enabled on agent	This specific agent can use the MCP integration
Default connection	Runtime chooses the user's default account unless another connection is pinned

See Enable MCP on agents.

Knowledge

Open Knowledge when the agent should answer from notebooks.

Knowledge notebooks must be enabled on the agent before the agent can search them.

Option	What it does
Search notebooks...	Finds notebooks by name or description
Notebook switch	Enables or removes that notebook for the agent
Enabled count	Shows how many notebooks the agent can search
Go to Knowledge	Opens Knowledge when no notebooks exist

Upload alone is not enough. The notebook must be enabled here.

Skills

Open Skills when the workspace has Skills enabled.

Skills add packaged knowledge or executable capabilities when the workspace enables them.

Option	What it does
Search skills...	Finds available skills
Type filter	Filters skills by type
Knowledge skills	Add context and expertise
Executable skills	Add actions the agent can use
Skill details	Shows the skill description and metadata before enabling

If Skills are disabled, the panel says Skills are not enabled in this workspace and tells you to contact an administrator.

Recommended configuration order

Profile & instructions
AI settings
Knowledge if the agent answers from documents
Tools and MCP integrations if the agent takes action
Objects if the agent works with structured records
Skills if your workspace uses them
Dynamic discovery review before sharing
Test in Build