How It Works

Your Salesforce Org Is Unique.
Your AI Agent Should Be Too.

A technical deep-dive into how REDA AI's extensible agent architecture reads your schema, follows your business logic, enforces your security model, and deploys across 4 channels with no-code simplicity — all from inside your own org.

The Core Problem

Generic AI Doesn't Know Your Org.

Every Salesforce org is a uniquely built system — custom objects, bespoke logic, carefully crafted security rules. Off-the-shelf AI sees none of it.

Salesforce Admin

Your Custom Schema Is Invisible

You've built custom objects, junction tables, and bespoke field structures over years. Generic AI tools know nothing about them — they operate on standard objects only, making them useless for your actual business data.

"We spent three months trying to connect an external AI to our org. It never understood our data model."
Process Owner

Your Business Logic Is Ignored

Your approval thresholds, escalation rules, and validation logic encode years of operational decisions. An AI that doesn't follow these rules doesn't just produce wrong answers — it creates immediate compliance risk and destroys user trust.

"The AI gave a customer a discount our policy doesn't allow. It had no idea the rule existed."
CTO / VP Engineering

A Second Security Model to Govern

Bolted-on AI means a separate system to secure, a separate vendor to audit, and a separate data pipeline to maintain. Every FLS rule you enforce in Salesforce has to be duplicated — and duplication creates gaps.

"We had to build a whole middleware layer to prevent the AI from seeing data users weren't supposed to access."
The Core Architecture

Three Levels of Salesforce AI Agent Extensibility

REDA AI isn't a fixed product — it's an extensible platform. At every level, you control more of what the agent knows, how it behaves, and what it can do. No level requires the next. Start at Level 1 and extend only as far as your use case demands.

1
Persona &
Context

Agent Persona & Context

Define who the agent is and what it sees on every interaction

Every REDA AI agent starts with an identity and an awareness of the record it's working within. You define the agent's purpose, tone, and role in plain English. Context fields load the most relevant Salesforce data automatically when a user opens a record page — so the agent always starts informed, never generic.

Agent PersonaUp to 5,000 characters. Define role, tone, boundaries, and purpose in plain language.
Context FieldsAuto-load any standard or custom field on record open. Priority hierarchy: Agent+RecordType → Agent → Universal.
Suggestion ChipsPre-built prompt shortcuts surface the most common user actions — reducing time-to-answer and improving consistency.
4 Deployment ChannelsAssign each agent to Internal, External Widget, Experience Cloud, or API/Flow — or all four simultaneously.
For AdminsConfigure the agent's identity and awareness through point-and-click setup. No code. Changes take effect immediately — no deployment cycle.
For CTOsThe context field priority hierarchy scales cleanly across multiple record types. One agent definition can serve an entire org without configuration sprawl.
2
Agent
Knowledge

Agent Knowledge — Custom Prompts

Five prompt types that give the agent deep understanding of your business

Beyond persona, REDA AI supports five structured prompt types assembled hierarchically at inference time. Together they teach the agent your org's data structures, behavioral rules, safety guardrails, and operational context — all written in plain English, all configurable without touching code.

Persona OverrideOverride the base persona for specific agents or contexts — e.g. a stricter compliance agent vs. a friendly sales assistant.
Custom InstructionsBehavioral rules, tone guidelines, escalation conditions, and process-specific directions the agent must always follow.
Schema PromptsDescribe your custom objects, field meanings, and picklist values in plain English so the agent reasons accurately about your data.
Action PromptsTeach the agent when and how to invoke tools — conditions for creating a task, when to send an email, how to structure a case update.
Guardrail PromptsHard limits on what the agent can discuss or do — topics to avoid, data it must never expose, actions requiring escalation.
Sort Order & A/B TestingPrompts assemble in priority order. Activate and deactivate independently to test changes without breaking live agents.
For AdminsWrite behavioral rules in plain English. The AI-powered Prompt Builder helps you write and refine prompts — no prompt engineering expertise required.
For CTOsFive layered prompt types provide granular control over behavior, knowledge, and safety. A/B testing supports safe iteration in production without full redeployment.
3
Agent
Ability

Agent Ability — Tools & Actions

What the agent can do — from pre-built actions to fully custom tools

Agent Ability defines what your agent can actually execute. REDA AI ships with 12 pre-built tools covering the most common Salesforce operations. Beyond that, Salesforce Flow integration connects agent decisions to your existing automations. For deeper programmatic control, agents can be invoked directly from Apex via the REDA AI Apex API. And when your use case needs something entirely bespoke, admins can build custom tools — REDA AI generates the Apex code automatically via Tool Builder.

12 Pre-Built Tools

No-code. Assign to any agent instantly.

QueryGet Record SearchCreate UpdateDelete Send EmailCreate Task Log EventSchedule SummariseDraft

Flow Integration

Declarative. Connect AI to your existing automation chains.

Invoke from conversation Trigger agent from Flow Unattended automation Admin-defined run context

Apex API

Programmatic. Invoke agents from any Apex context.

Apex triggers Batch jobs Custom controllers Deep org integration

Custom Tools

Code-optional. REDA AI writes the Apex — you extend it.

Tool Builder (clicks) AI-generated Apex Custom logic extension Org-specific integrations
For AdminsAssign pre-built tools with a click. For custom tools, describe what it should do and Tool Builder generates the Apex. No Apex knowledge required to start — Apex skills unlock further customisation.
For CTOsThe Apex API enables deep integration with existing org logic without a UI or Flow. All tool actions require user confirmation before execution. Custom tool code stays inside your org — no external execution.
AI-Assisted Configuration

No-Code: AI Builds Your
Salesforce AI Agent

The fastest path from an idea to a live agent is letting REDA AI do the heavy lifting. Click "Build Agent with AI" and REDA AI analyses your org's schema, your record types, and your existing field configuration — then generates a fully configured agent in under two minutes.

1
Schema AnalysisREDA AI reads your org's metadata — objects, fields, relationships, and picklist values relevant to the agent's purpose.
2
Prompt GenerationGenerates an optimised persona, schema prompts that describe your custom objects accurately, and guardrails tailored to the deployment channel.
3
Context Field RecommendationsRecommends the highest-signal context fields for the target object — the data most likely to produce accurate, relevant responses.
4
Suggestion Chip GenerationCreates ready-to-use suggestion chips based on the agent's role and common user actions on that record type.
1–5 credits per analysis. Run it as many times as you need.
Build Agent with AI — Opportunity Agent
Analysing org schema…
Objects foundOpportunity, Quote, Contract + 4 custom
Custom fields47 analysed, 12 recommended
RelationshipsAccount, Contact, Opp mapped
Generated persona
RoleSenior Deal Intelligence Advisor
ToneAnalytical, concise, action-oriented
FocusDeal risk, next steps, stakeholder gaps
Context fields (priority order)
1. AmountHigh signal — confirmed
2. Close DateHigh signal — confirmed
3. StageHigh signal — confirmed
4. Deal_Risk__cCustom field — confirmed
Suggestion chips
Summarise this deal Identify risks Draft follow-up Next best action
✓ Agent ready to deploy
Technical Architecture

How a Request Works — Inside Your Org

Every REDA AI interaction follows the same four-step lifecycle. Security is enforced at every step — not bolted on at the end.

1

User Sends a Message

The message originates inside Salesforce's authenticated session. On the Internal channel, the full Salesforce security context is already established — no re-authentication, no token exchange.

Session-authenticated
2

Context Assembled Securely

REDA AI loads context fields for the current record — but only fields the authenticated user can see. FLS, sharing rules, and object permissions filter the data before assembly. No raw records leave the org.

FLS + Sharing enforced
3

AI Processes the Request

The assembled context — your custom prompts, FLS-filtered field data, and the user's message — is sent to the AI engine. No raw CRM records are included. The AI reasons within the guardrails you've defined.

Zero raw record egress
4

Response & Action Delivered

The response is returned inside the Salesforce UI. Any write actions require explicit user confirmation via Action Confirmation Cards before execution. A full audit trail is created automatically as standard Salesforce records.

Confirmed actions + audit trail

For Technical Evaluators — What Never Leaves Your Org

Raw Salesforce records are never transmitted. The AI engine receives only assembled, FLS-filtered context — field values the running user is authorised to see, structured as described in your schema prompts. No external database stores this data. No inference result is persisted outside your org. Every interaction is logged as a standard Salesforce record — queryable, reportable, and auditable with the tools you already have.

Enterprise Observability

Measure, Optimise, and Govern Every Interaction

Deploying AI without observability is flying blind. REDA AI gives you full visibility into cost, quality, and compliance — from inside your Salesforce org.

REDA AI Analytics — Dashboard
419
Total Conversations
3.3k
Credits Available
6
Scheduled Actions
47
Open Incidents
Conversations by Channel
419
Total
Web Slack Portal Flow
Incident Trend
Oct '25 Nov Dec Jan '26 Feb
Credit Usage by Period
Oct
Nov
Dec
Jan
Feb
Credits used

Credit Efficiency Ratings

Every conversation is scored automatically — Optimal, Good, Needs Review, or High Cost. Identifies which agents and interaction patterns consume disproportionate credits so you can optimise proactively.

Per-Message Token Breakdown

See exactly where credits are consumed at the message level — system prompt weight, context field volume, response length. Color-coded intensity indicators make patterns immediately visible.

Actionable Optimisation Suggestions

REDA AI analyses consumption patterns and surfaces specific recommendations — reduce tool count, add a schema prompt, tighten context fields — with estimated credit savings before you apply each change.

Incident Logging & AI Debugging

When an agent behaves unexpectedly, the incident is logged automatically with full conversation context. AI-powered debugging identifies the likely prompt or configuration root cause and suggests a fix.

12-Month Credit History

Full credit usage history across all agents and channels. Track consumption trends, forecast annual spend, and identify seasonal patterns — all from a standard Salesforce dashboard your team already knows.

Compliance-Ready Audit Trail

Every conversation, action confirmation, and tool execution is a standard Salesforce record. Compatible with any SIEM, reporting tool, or compliance framework your organisation already uses — no additional connectors needed.

Optimal — Low cost, high quality
Good — Within expected range
Needs Review — Above avg. consumption
High Cost — Optimisation recommended
Deployment

From Zero to Live Agent in Under an Hour

No developers. No middleware. Four steps — each one shorter than you'd expect.

01

Install the Managed Package

Find REDA AI on the Salesforce AppExchange and install directly into your org. One-click deployment — no external infrastructure, no middleware, nothing outside Salesforce.

Under 5 minutes
02

Activate Your Licence

Open the REDA AI Admin tab inside Salesforce and enter your licence key. The Trial plan activates instantly — 500 credits, one agent, no credit card required.

30 seconds
03

Build Your First Agent with AI

Open Agent Builder, describe the agent's purpose, and click "Build Agent with AI." REDA AI reads your org's schema and auto-generates the persona, prompts, context fields, and suggestion chips. Review and deploy.

Under 10 minutes
04

Deploy to Any Channel

Drag the Lightning component onto any record page. Embed the External Widget on your website. Add to Experience Cloud portals. Invoke from a Salesforce Flow or directly from Apex for unattended automation.

Minutes per channel

Your first agent can be live in under an hour.

No professional services needed to get started. Implementation support available for advanced configurations and custom tool development.
Start Free — 500 Credits →
In Practice

Three Agents. Three Channels. Real Scenarios.

Not generic department labels — actual deployment patterns with the tools and channels each one uses.

Deal Intelligence Agent

Sales team · Opportunity record pages

What are the biggest risks with this deal right now?
Three risks flagged: Close date is 4 days away but no signed quote on file. Economic buyer (CFO) not contacted in 31 days. Competitor mentioned in last call notes — no counter-strategy logged.

Want me to draft a CFO outreach email and create a follow-up task?
Yes, do both.
Draft ready for your review. Task created for Monday 9am. Confirm to send the email?

Customer Support Agent

External widget · Public website

My invoice shows a different amount than what I was quoted.
I can help with that. Could you share your account email or invoice number? I'll pull up the details and walk you through what changed — and raise a case with our billing team if needed.
Account email is sarah@example.com
Found your account. Invoice #INV-2847 shows a mid-cycle rate adjustment applied Feb 14. I'm opening a billing case now — you'll receive email confirmation within the hour.

Automated Lead Enrichment

Apex API · Unattended, triggered on Lead creation

Apex trigger fires: New Lead created → REDA AI Apex API invokes agent automatically — no UI, no user interaction
Lead analysed: Company size inferred from domain (enterprise, 2,000+ employees). Industry: FinTech. Job title maps to economic buyer persona. Recommended segment: Enterprise Financial Services.
Actions executed: Lead.Segment__c updated → "Enterprise FinServ". Routing rule triggered → assigned to Enterprise queue. Follow-up task created for assigned rep.
Start Building

Ready to Deploy Your First Agent?

500 free credits. Your real Salesforce org. No developers, no credit card, no consultants required to get started.

No credit card.  ·  No commitment.  ·  100% Salesforce-native.

How REDA AI Compares to Agentforce

Unlike Agentforce, REDA AI uses a 3-level extensible agent architecture (Persona → Knowledge → Ability) that deploys as a 100% Salesforce-native managed package — no Data Cloud dependency, no per-conversation fees, and no external data transmission. Where Agentforce requires Data Cloud and charges $2 per conversation, REDA AI uses credit-based pricing starting at $1,500/org/month with a free trial. REDA AI is built for the majority of Salesforce orgs that need production-ready AI agents without the overhead, complexity, and cost of Agentforce.

Last updated March 2026