Configuring Agent Settings
AI funding and usage
AI included uses REDA credits. Bring your own AI uses your organisation’s provider account, billed directly by that provider. The licence selects one funding model for the organisation. In either model, REDA AI governs configured tools and actions in Salesforce. Set up your licence and AI accounts.
With AI included, select the agent’s Processing Tier. With Bring your own AI, select AI Connection and Model in the agent’s AI Account section, or inherit the organisation default. Different agents can use different supported providers and models. Prompts, tools, record context and channel permissions still define each agent’s job. Configure the agent’s AI account.
Core Agent Properties
| Property | Type | Description |
|---|---|---|
| Display Name | Text | The name shown to users in the chat UI. Make it descriptive and action-oriented (e.g., "Sales Pipeline Helper" not "Agent 1"). |
| Channel | Picklist | Determines where the agent is deployed and how it authenticates. Options: Internal, External Widget, Community, API. |
| Active Status | Checkbox | When unchecked, the agent is disabled. Inactive agents show a "not available" message and do not respond. |
| Persona | Long Text Area (5000) | The opening system prompt that defines the agent's role, tone, and purpose. Example: "You are a friendly sales assistant helping close deals..." |
| Priority | Number | When multiple agents appear in the same location, the agent with the highest priority number is displayed. Packaged default: 50. |
Message & Token Settings
| Property | Type | Purpose |
|---|---|---|
| Max Message Length | Number | Maximum characters a user can send in a single message. Packaged default: 2000 characters. Use this to prevent extremely long prompts that consume tokens. |
| Max Session Messages | Number | Maximum number of messages per session before the session resets and conversation history is cleared. Packaged default: 50. |
| Session Timeout (Minutes) | Number | How many minutes of inactivity before a session expires. Default: 30 minutes. Resets every time the user sends a message. |
| Prompt Complexity Budget | Number | Caps how much of the agent’s system prompt is assembled, not how much the AI may use in a turn. The assembly engine concatenates Custom Prompt records in Sort Order until the budget is reached, and the Persona text counts toward it. Packaged default 8,000. The field is labelled “Prompt Complexity Budget” on the record — older material calls it Token Budget. Lowering it does not make each answer cheaper; it silently drops the prompts that no longer fit. |
| Use Streaming | Checkbox | Off by default. With it on, every AI call for this agent streams, which keeps the HTTP connection alive as tokens arrive and prevents read timeouts. Recommended for agents that process files, do complex reasoning, or produce long answers. If REDA AI ever tells you to “enable streaming on the agent to avoid timeouts”, this is the checkbox it means. |
| Start On Load | Checkbox | Off by default. When on, the chat component sends a greeting the moment it mounts on a record page, prompting the agent to review the record and lead with insights, action items or risk flags — instead of waiting to be asked. This is what turns the panel from a search box into a proactive assistant. |
| Conversation Window (Hours) | Number | A positive value sets the automatic-resume window in hours. Blank uses the organisation default, initially 30 minutes; 0 disables automatic resume. Eligible conversations must match the channel, agent, user and record context required by that channel. See the conversation-continuity section below. |
Performance & Cost Settings
These settings control what the agent costs to run. Every one of them is optional: leave it blank and the agent keeps using the organization-wide default, so an agent you never touch behaves exactly as it did before. They appear in the Performance & Cost Controls section of the Agent Definition layout, except where noted.
| Property | Type | Purpose |
|---|---|---|
| Processing Tier | Picklist | Standard, Efficient or Economy. Blank behaves as Standard.
|
| Economy Tier Consent | Checkbox | Required before using Economy. Review the displayed data-processing terms, then select this checkbox only when your organisation agrees. Without consent, Economy requests are refused; REDA AI does not silently switch to another tier. |
| Enhanced Caching | Checkbox | Keeps the agent's stable context (its instructions and tool list) warm for one hour instead of five minutes, so a conversation resumed after a pause is not charged as though it had just started. Turn it on for agents people chat with; leave it off for autonomous agents that finish in one go. It can never cost you more — REDA absorbs the higher storage cost. |
| Max Output Tokens | Number | Caps how long one reply from this agent may be. Blank uses the organization-wide default. Not to be confused with Token Budget above, which caps what is sent to the AI; this caps what comes back. |
| History Token Budget | Number | How much earlier conversation is re-sent on every turn so the agent remembers what was said. Blank uses the organization-wide default. Usually the largest single cost lever on a chatty, high-volume agent — but lower it in steps, because too low and the agent starts forgetting earlier context. |
| Web Search Max Uses | Number | How many web searches this agent may run per turn. Blank uses the organization-wide default. Only meaningful when Allow Web Search is enabled; each search costs credits beyond the usual token charge. |
| Max Continuation Depth | Number | How far the agent may keep working across turns on the external and community channels. Blank uses the organization-wide default; 0 disables continuation entirely. Sits in the External Controls section, because it applies only to those channels. |
| Execution Context | Picklist | User (default) or System. Applies only to API-channel agents and is ignored everywhere else. System lets an unattended agent bypass object and field permissions — it does not let it see records it could not already see. Sits in Context & Persona, directly below Channel. See Data Access & Permissions before changing it. |
Tool & Feature Settings
| Property | Type | Description |
|---|---|---|
| Tools Mode | Picklist | Always: load tools on every page (high token usage). Record Pages Only: load tools only on record pages (saves tokens on home/list pages). |
| Max Tools | Number | How many tools this agent offers the AI per turn (1–100). Blank behaves as 25. Every tool the agent can reach is described to it on every message, so a long list costs credits each turn whether the tools get used or not. Trim it on an agent that only ever needs a handful. |
| Max Tool Iterations | Number | How many steps the agent may take within a single turn when chaining tools together (3–25). Blank behaves as 6. Raise it for genuinely multi-step work; lower it to stop an agent looping on a task it cannot finish. This replaces the organization-wide setting of the same name, which is no longer used. |
| Deep Reasoning | Picklist | How thoroughly the agent reasons before answering: Off (default), Low, Medium, High, Xhigh, Max. Higher levels improve accuracy on complex questions at the cost of slower replies and more credits. A blank value is treated as Off. See Deep Reasoning. |
| Allow Automated Requests | Checkbox | When enabled, Salesforce Flows and automations can invoke this agent programmatically. Required for Flow integration. |
| Default Context Objects | Text | Comma-separated list of Salesforce objects (e.g., "Account, Opportunity, Contact") this agent primarily works with. |
| Allow Web Search | Checkbox | When enabled, the agent can search the web for real-time information during conversations. Each web search consumes a small number of additional credits beyond the base conversation cost. Default: disabled. See Web Search. |
| Allow File Upload | Checkbox | When enabled, users can upload files (PDF, CSV, images, text, JSON) into conversations. The agent reads and processes the file content. Default: enabled. See File Upload. |
| Supported File Types | Text | Comma-separated list of file extensions this agent accepts (e.g., "pdf,csv,txt"). Leave blank to use the global Allowed File Extensions setting from Platform Settings. |
| Auto-Execute File Actions | Checkbox | When enabled, the agent performs file-based actions (e.g., CSV import) immediately without asking for user confirmation. Default: disabled. |
Agent Collaboration Settings
These control whether this agent can take part in agent-to-agent delegation — one agent consulting another mid-conversation. Both switches start off, so an upgrade never changes how an existing agent behaves.
| Property | Type | Description |
|---|---|---|
| Agent Invocable | Checkbox | Lets other agents consult this one. Off by default. It is not enough on its own — Allow Automated Requests must also be on, because a consultation arrives as an automated request. If only one is ticked, the calling agent receives a message naming the one that is missing rather than failing silently. |
| Max Agent Call Depth | Number | Set on the agent that asks. How many levels deep consultations may chain: 1 (the default) means it may consult one agent, and that agent cannot consult a third; 0 means never consult. Blank behaves as 1. Each level adds response time and credits, so raise it only for a case you have seen need it. |
External Widget Settings
These settings apply when the Channel is set to "External Widget" (agent deployed on a website).
| Property | Type | Description |
|---|---|---|
| Allowed Origins | Text | Comma-separated list of allowed website domains (e.g., "https://example.com, https://www.example.com"). Required for External Widget. Prevents unauthorized embedding. |
| External Welcome Message | Long Text Area | Welcome message displayed when the External Widget chat opens. Example: "Hi! How can I help you today?" |
| Start on Load | Checkbox | When enabled, the External Widget chat window opens automatically when the page loads. When disabled, users must click a button to open the chat. |
Appearance Settings
| Property | Type | Description |
|---|---|---|
| Icon Name | Text | SLDS (Salesforce Lightning Design System) icon name displayed next to the agent name. Packaged default: "standard:bot". Example: "standard:service_resource". See SLDS icons documentation. |
| Mascot Image | File / Image | Optional custom mascot image or avatar for the agent. Displayed in the chat UI. Recommended: 200x200px or larger. |
Rate Limiting Settings
| Property | Type | Description |
|---|---|---|
| Rate Limit per Hour | Number | Maximum number of interactions (messages) a single user can send to this agent per hour. Packaged default: 500. |
| Rate Limit per Session | Number | Maximum number of interactions a single user can send in one session. Packaged default: 10. |
Data Management Settings
| Property | Type | Description |
|---|---|---|
| Archival Retention (Days) | Number | How many days to keep this agent's conversations before they are automatically archived. Blank uses the org-wide default of 180 days (configurable in Platform Settings). The minimum per agent is 60 days. |
| Skip Archival | Checkbox | When enabled, conversations with this agent are exempt from automatic archival. Use for agents handling sensitive data that should be retained longer. |
Configuration Workflow
- Open Agent Definition — Navigate to the Agent Definition record you want to configure.
- Review Core Settings — Ensure Display Name, Channel, and Persona are correct.
- Set Performance Limits — Configure Max Message Length, Max Session Messages, and Token Budget based on your use case.
- Enable Automations — Turn on Allow Automated Requests if you plan to invoke the agent from Flows.
- Configure Channel-Specific Settings — If using the External Widget channel, set Allowed Origins and a welcome message.
- Add Appearance — Optionally set an Icon Name and Mascot Image for branding.
- Set Rate Limits — Configure rate limits if you want to prevent abuse or overuse.
- Test — Save changes and test the agent in a sandbox environment.
Performance Tuning Tips
The Token Budget setting directly affects cost. Lower token budgets mean cheaper interactions but may limit the AI's reasoning depth. The packaged default is 8000 tokens per turn — lower it in steps if cost per interaction matters more than headroom, and monitor for responses getting cut off.
Use Record Pages Only for Tools Mode — If your agent is deployed on a record page in Salesforce, set Tools Mode to "Record Pages Only" to avoid loading tools on home pages and list views, saving tokens and cost.
Set Reasonable Session Timeouts — A Session Timeout of 30 minutes is standard for most use cases. Shorter timeouts save memory; longer timeouts preserve conversation context.
Use Max Session Messages for Long Conversations — If you notice conversations getting too long and token-heavy, set a Max Session Messages limit (e.g., 50) to reset the conversation after that many messages.
Channel-Specific Configuration
For Salesforce users. Set Tools Mode to "Always" or "Record Pages Only", enable tools freely, and assign a persona tailored to your team's needs.
For website visitors. Set Allowed Origins, configure an External Welcome Message, and be conservative with tool assignments (only tools that make sense for guest users).
For portal users. Tools and context fields are limited by the portal user's permissions. Configure securely to prevent unintended data access.
Common Configuration Scenarios
| Scenario | Recommended Settings |
|---|---|
| Sales Team Assistant (Internal) | Tools Mode: Always · Token Budget: 2,500 · Automated Requests: enabled · Session Timeout: 45 min |
| Website Chat Bot (External Widget) | Start on Load: enabled · Allowed Origins: your-domain.com · Token Budget: 1,500 · Rate Limit per Hour: 10 |
| Customer Portal Agent (Community) | Tools Mode: Record Pages Only · Token Budget: 2,000 · Max Session Messages: 30 · Skip Archival: disabled |
| API Integration Agent (API) | Automated Requests: enabled · Token Budget: 3,000 · Rate Limit per Hour: 100 |
Understand the Economy credit change
Economy ordinary usage now applies a 0.40 multiplier to the base credit rate. For unchanged token quantities and base rates, an example costing 100 base credits costs 40 Economy credits. At the previous 0.10 multiplier it cost 10: the new debit is four times the previous Economy debit, and 60% below base. This changes credit consumption, not the subscription price. File pre-extraction is charged separately at full rate. A provider fallback can use a different effective tier; inspect the actual transaction rather than applying 0.40 to every mixed operation. New credit transactions record Processing Tier; historical rows are not backfilled. Economy still requires consent and compatible licence/provider configuration; a missing consent is refused, not silently processed at another tier.
Review conversation continuity
A blank Conversation Window (Hours) now uses the organisation default, initially 30 minutes. A positive value sets the agent's window; 0 disables automatic resume. Matching remains channel-specific: API uses agent and record context; authenticated Internal/MCP uses agent, user and the relevant record context. Community uses the shared window setting instead of the external-session timeout. Home, app-page and utility-bar chat can resume eligible conversations. Explicit conversation identifiers remain subject to their route's access checks. Review agents that relied on blank meaning a new conversation, and use 0 where that behaviour is required. Packaged permission fixes also address Internal and Community conversation access.
Frequently Asked Questions
If the AI's response would exceed the token budget, REDA AI truncates the response or returns a partial answer. Set the budget high enough for your use case. Monitor token usage in logs and adjust upward if responses are consistently cut off.
Yes, but changing the channel may break existing deployments. Internal agents deployed on a page will not work if you change to the External Widget channel. Plan your channel choice carefully before deploying.
Max Session Messages limits the number of turns in a conversation. Session Timeout ends a session if the user is inactive. Both reset the conversation context — use them together to manage token budget and conversation freshness.
No. Allowed Origins is only used for the External Widget channel. Internal agents authenticate via the Salesforce session and don't need origin restrictions.
Next Steps
After configuring agent settings, learn how to activate and deactivate agents, or explore prompt management to refine your agent's behavior.