Smart Agent (card)
The Smart Agent card is the central configuration page for a single AI agent.
Merged page: This file covers two pages displayed together on the agent card.
Page 72778301 "Smart Agent Card QUA"(Card) —Page.72778301.SmartAgentCard.alPage 72778309 "SA Agent Image FactBox QUA"(CardPart factbox) —Page.72778309.SAAgentImageFactBox.alSource table:
Smart Agent
The Smart Agent card is the central configuration page for a single AI agent. You define the agent's identity, choose its AI model and autonomy level, enable capabilities such as web search or knowledge base access, write its instructions, and manage its email channel settings. The Agent Image factbox on the right lets you upload a photo or icon for the agent.
In plain terms
This is the single most important page in the app — it's where you build an agent. An agent is a configured AI assistant your users chat with to get work done in BC.
Everything that defines one agent lives on this single card. Just as a permission set plus user setup together decide what a person can see and do, this card bundles the agent's identity (name, icon), its brain (the AI model), its permissions (capabilities, data connections, tools), and its standing orders (instructions).
A few groups on this page use AI ideas worth knowing before you start:
- Instructions are the agent's always-on standing orders — its role, rules, and tone. The AI re-reads them before every message, so you never restate them. This is what AI people call the system prompt — much like the defaults baked into a posting group, they apply automatically to everything the agent does without anyone restating them.
- Capabilities (BC Data Access, Knowledge Base, Web Search, Image Analysis) are coarse on/off gates, like the top-level toggles on a permission set. Turning one on only opens the door — the detail pages (Tool Configurations, Knowledge Files) decide what's actually behind it.
- Autonomy Level controls how often the agent stops to ask before acting — from "approve every step" to "full autopilot".
- Thread Memory is a per-conversation scratchpad where the agent tracks its goal, plan, and findings, so it doesn't lose the thread on long tasks.
Worked example. A "Finance Helper": Name = Finance Helper, AI Model = Smart, Autonomy Level = Medium, BC Data Access = on (connected to Customers and Sales Invoices), Instructions = "You help finance chase overdue invoices. Always show amounts in the customer's currency; never modify records without confirmation." You Activate it, then a user asks "Who's more than 30 days overdue?" and it answers from live BC data.
New here? Read the end-to-end build walkthrough in the primer first.
How to open it
- From the Smart Agents list: select an agent and choose Edit, or choose New Agent to create one.
- Drilling down on any agent row in the Smart Agents list also opens this card.
Fields
Identity
| Field | Type | Description |
|---|---|---|
| No. | Integer | The unique number identifying this agent. Read-only. (shown under Show more) |
| Name | Text | The display name users see in the chat sidebar and in Teams. Required. |
| Description | Text | A short description of what this agent does. Shown to users when picking an agent. |
| Icon | Text | A short visual identifier for this agent, e.g. "FIN", "INV", "CSR". |
| Department Code | Code | Optional department used to group this agent in the chat sidebar tree. Links to Smart Agent Departments. |
| Status | Enum | Only Active agents can be used for chat. Use Activate / Deactivate actions to change. Values: Draft, Active, Disabled. Read-only. |
Configuration
| Field | Type | Description |
|---|---|---|
| AI Model | Text | Select the AI model for this agent from the backend-managed model list. Required. Use the lookup to pick from the available model tiers. |
| Autonomy Level | Enum | Controls how independently the agent acts. Values: Low: Step by Step — asks permission for every action. Medium: Plan and confirm — asks clarifying questions upfront, then acts. High: Confirm write tool calls — only confirms destructive actions at the final step. Autonomous: Act and report — full autopilot, no questions. |
| Output Type | Enum | Chat = natural language for users. JSON = structured output for programmatic callers (requires a JSON Schema). Values: Chat, JSON. |
| Backend Assistant ID | Text | The backend reference for this agent. Populated automatically on activation. Read-only. (shown under Show more, visible only when Show Advanced Options is on in Setup) |
How to choose — AI Model: start on Smart (mid-range). Go higher (Expert/Premium) only if the agent makes reasoning mistakes on real tasks; drop to Fast for simple lookups to save credits. See AI Models.
How to choose — Autonomy Level: think of it like the number of approval steps before posting. New agents and anything that writes data → Low or Medium while you build trust. Read-only reporting agents → High or Autonomous is fine. Autonomy is the general posture; per-tool On Confirmation settings are still enforced on top of it.
How to choose — Output Type: leave it on Chat (a reply a person reads) for almost everything. Pick JSON only when another system or automation needs structured, machine-readable output instead of a human — the equivalent of an API payload rather than a printed report.
Capabilities
| Field | Type | Description |
|---|---|---|
| BC Data Access | Boolean | Let this agent read and write Business Central data. Configure specific objects under BC Data Connections. |
| Image Analysis | Boolean | Let this agent analyze uploaded images. |
| Knowledge Base | Boolean | Let this agent reference uploaded documents. Manage files under Knowledge Files below. |
| Web Search | Boolean | Let this agent search the web for current information. |
Memory & Access
| Field | Type | Description |
|---|---|---|
| Thread Memory (chars) | Integer | Maximum characters for the agent's per-session scratchpad. 0 = disabled. The agent uses this to persist plans, verified data, and task context across turns. |
| Allow User to Edit Thread Memory | Boolean | When on, users can edit Goal, Plan, Verified Data, Decisions & Rules, and Errors & Retries sections from the chat side panel during the conversation. Only available when Thread Memory is enabled (greater than 0). |
| Allow All Users | Boolean | When on, every user in the tenant can see and use this agent. When off, only users in Authorized Users can access it — if nobody is listed, the agent is hidden from everyone. |
Email Channel
This group is visible only when Output Type is Chat.
| Field | Type | Description |
|---|---|---|
| Enable Email Channel | Boolean | Let this agent monitor an email inbox and auto-reply. Requires Graph API credentials to be configured in Smart Agents Setup. |
| Monitored Mailbox | Text | The Microsoft 365 email address this agent monitors. Required when Enable Email Channel is on. |
| Check Interval (Minutes) | Integer | How often to check for new emails. Minimum 1 minute, default 5. |
| Auto Reply | Boolean | Automatically send the AI-generated response as an email reply. |
| Validate Sender As | Enum | Look up sender in a BC table. Matched senders scope tool calls; unrecognised senders get a rejection reply. Values: None, Customer, Vendor, Employee, Salesperson. |
| Last Checked | DateTime | When this agent last checked the mailbox. Read-only. (shown under Show more) |
Instructions
| Field | Type | Description |
|---|---|---|
| Instructions (rich text editor) | Blob | Detailed instructions for how the agent should behave. Use the rich text editor for formatting. |
JSON Schema
This group is visible only when Output Type is JSON.
| Field | Type | Description |
|---|---|---|
| Schema Preview | Text | Preview of the uploaded JSON Schema. Use Import / Export / Clear actions to manage it. Read-only. |
BC Data Connections
This subpage is visible only when BC Data Access is enabled. It lists the Business Central objects (tables, APIs) the agent is allowed to read or write.
Knowledge Files
This subpage is visible only when Knowledge Base is enabled. See Knowledge Files for the full field and action reference.
Usage Statistics
This group is visible only after the agent has been saved (No. is assigned).
| Field | Type | Description |
|---|---|---|
| Last Used At | DateTime | When this agent was last used. Read-only. |
| Total Credits Used | Decimal | Total credits consumed by this agent. Read-only. |
| Current BC Environment | Text | The current Business Central environment for this session (not stored per agent). Read-only. (shown under Show more) |
| Current BC Company | Text | The current Business Central company for this session (not stored per agent). Read-only. (shown under Show more) |
Agent Image (factbox)
| Field | Type | Description |
|---|---|---|
| Image | Media | The image for this agent. Use Import to upload a new image. |
Actions
Process
| Action | What it does |
|---|---|
| Activate Agent | Set this agent to Active so it can be used for chat. Provisions the agent on the backend. Shows a progress dialog; validates instructions and syncs to the backend before activating. |
| Update Agent | Push the latest settings (instructions, capabilities, model) to the backend. Available only after the agent has been provisioned. |
| Reset to Default | Restore this agent to its built-in default definition when one exists. Overwrites instructions and template settings, but keeps usage history and backend identity. |
| Deactivate Agent | Set this agent to Disabled so it can no longer be used for chat. Prompts for confirmation; unpublishes from Teams if currently published there. |
| Manage Shares | Invite other tenants to share this agent, or revoke existing shares. Requires the agent to be provisioned on the backend. |
| Delete Agent | Permanently delete this agent and all its data (sessions, messages, files, data sources). Also removes it from the backend. This cannot be undone. |
| Chat Now | Open a chat session with this agent. Available only when the agent is Active and its Output Type is Chat. |
| Chat in Teams | Open a chat with this agent in Teams. The Teams app must be installed first (use Download Teams App). Available only when the agent is published to Teams. |
| Download Teams App | Download the Teams app manifest ZIP for this agent. Upload it in Teams via Apps > Upload a custom app. Available only when the agent is published to Teams. |
| Publish to Teams | Publish this agent to Microsoft Teams so users can chat with it directly in Teams. Available only when the agent is Active. |
| Unpublish from Teams | Remove this agent from Microsoft Teams. Available only when the agent is published to Teams. |
| Email Log | View the email processing log for this agent — incoming emails, replies, and errors. Available only when Enable Email Channel is on. |
| Setup Email Job Queue | Create or update the recurring Job Queue Entry that polls for incoming emails. Required for email integration to work. Available only when Enable Email Channel is on. |
| Check Emails Now | Manually trigger an email check for this agent right now. Useful for testing email integration. Available only when Enable Email Channel is on and the agent is Active. |
| Import JSON Schema | Import a .json file containing the JSON Schema that defines this agent's output structure. Visible only when Output Type is JSON. |
| Export JSON Schema | Download the current JSON Schema as a .json file. Visible only when Output Type is JSON. |
| Clear Schema | Remove the current JSON Schema from this agent. Visible only when Output Type is JSON. |
| Authorized Users | Manage which users are allowed to chat with this agent. Only used when Allow All Users is off. Opens the Authorized Users page. |
| Suggested Prompts | Manage the suggested prompts shown to users at the start of a new chat. Opens the Suggested Prompts page. |
| Published Channels | See every channel this agent is published to (BC, Teams, etc.) with sync timestamps. Opens the Published Channels page. |
| Web Resources | Manage external web links this agent can reference. |
| Test Cases | Manage AI test cases for this agent. Each test replays a saved conversation and grades the result with the same model. |
| Chat Sessions | Browse all chat sessions for this agent and promote any past conversation into a test case. |
Related pages
- Smart Agents list — parent page
- Knowledge Files — embedded subpage (visible when Knowledge Base is enabled)
- Authorized Users — opened via the Authorized Users action
- Published Channels — embedded subpage and dedicated action
- Suggested Prompts — opened via the Suggested Prompts action
- Smart Agent Departments — linked via the Department Code field
Notes
- A newly created agent starts in Draft status. You must use Activate Agent before anyone can chat with it.
- Before activating, ensure your tenant is registered: go to Smart Agents Setup and click Register Tenant.
- If the agent's instructions conflict with system rules, Activate offers a suggested fix. You can accept it or cancel activation.
- When you switch Output Type from JSON back to Chat, any imported JSON Schema is automatically cleared.
- The Email Channel group requires Graph API credentials (Azure AD Tenant ID, Client ID, Client Secret) to be configured in Smart Agents Setup before you can enable email.
- Each Microsoft 365 mailbox can be monitored by only one agent. Setting the same email address on two agents is blocked.
- Enabling Image Analysis or Knowledge Base on an already-active agent immediately syncs the change to the backend. If the sync fails, the change is rolled back.
- Deactivating an agent also unpublishes it from Teams (with a confirmation prompt if the Teams unpublish call fails).
- Imported JSON Schemas larger than 10 KB trigger a warning about token cost before they are accepted.