The REDA AI Platform
The AI features described elsewhere in this section — invoice OCR, bank-statement matching — are built into REDA Financials itself. But they are not the whole story. REDA (the maker of REDA Financials) also builds REDA AI: a full AI agent platform for Salesforce, available as its own managed package. Because both products live natively in your Salesforce org, an organization running REDA Financials can add conversational, configurable AI agents that work alongside its accounting — with the same users, the same permissions, and no integration project.
What REDA AI is
In REDA's own words: "REDA AI is a Salesforce-native AI agent platform that installs as a managed package and runs entirely inside your org — no middleware, no Data Cloud, no parallel security model." You install it from the Salesforce AppExchange, an administrator configures agents in days rather than quarters, and users chat with those agents right where they work — on record pages, from the utility bar, in a customer portal, or through a chat widget on your public website.
The five ideas that make an agent
| Concept | What it is |
|---|---|
| Agent | The AI persona users talk to — it has a name, a deployment channel, tools, prompts, and context fields. An org can run several, each independent. Understanding Agents |
| Prompts | Written instructions that control behaviour — what the agent knows, what it may do, how it responds. Guardrail prompts carry your restrictions and safety rules. Understanding Prompts |
| Tools | Capabilities to read or act on Salesforce data. Read tools query and describe; Action tools create records, update them, send emails, create tasks — with the user's confirmation before anything changes. Understanding Tools |
| Context Fields | Fields the agent automatically reads from the record page the user is on, so answers arrive already grounded — subject to field-level security. Understanding Context Fields |
| Credits | The usage unit — each interaction consumes credits based on the text processed, allocated monthly by plan and tracked per conversation. How Credits Work |
What agents can do
- Answer and summarise — questions about your Salesforce data in real time; summaries of any standard or custom record.
- Act, with confirmation — create or update records, send emails, create tasks and events, schedule follow-ups; action tools show a confirmation card before executing.
- Read documents — agents process uploaded PDFs, images, and CSVs natively, extracting structured data into Salesforce records; an inbound email channel is purpose-built for document intake such as vendor invoices and supplier feeds.
- Run in the background — Salesforce Flows can invoke an agent as a drag-and-drop action for unattended work: automated processing, enrichment, routing, notifications. Heavy requests run asynchronously and post their reply back to the conversation.
- Speak your users' languages — one agent responds in each user's Salesforce language, without per-language configuration.
Where agents live
| Channel | Who uses it |
|---|---|
| Internal | Your Salesforce users — a chat panel on record pages, the home page, or the utility bar, reading record context automatically. Internal Channel |
| Community | Authenticated portal users in Experience Cloud, under their portal permissions. Community Agent |
| External Widget | Visitors to your public website, via a single script tag — no Salesforce login, guarded by domain whitelists, rate limits, and daily credit caps. External Widget |
| API / Flow | No human at all — Flows, Apex, and scheduled jobs invoke agents headlessly for automation. Flow Integration |
On top of these run email channels (inbound document intake and agent-sent outbound email), a messaging integration pattern for systems like WhatsApp and SMS via extension connectors, and voice input in REDA AI chat. The REDA AI KB documents each surface.
Security and governance — the part accountants ask about first
| Claim | What it means |
|---|---|
| "Your CRM data never leaves your Salesforce org." | No REDA-side server holds your records, no third-party database stores them; only the assembled context you configure is sent for AI reasoning, and the primary provider’s published terms exclude your data from model training. |
| "The agent only sees what the running user is allowed to see." | REDA AI inherits your field-level security, sharing rules, and profiles — no duplicate security model, no privilege escalation. If a user can't see a field, neither can their agent. |
| Confirmation before change | Action tools present a confirmation card — the action and its key values — before anything is written. |
| Everything on the record | Conversations and messages are stored as Salesforce records in your org, reportable with standard reports; incident logs capture errors for admins; archival and audit records are immutable even to admins. |
| Certifications | ISO 27001:2022 and SOC 2 Type II certified, GDPR compliant — detailed on the governance page. |
What this means next to REDA Financials
REDA Financials documents are Salesforce custom objects, and REDA AI agents are built to "work with any standard or custom Salesforce object." Install both packages in one org and the consequence is direct: agents can read and — with action tools and confirmation — write accounting records the same way they work with anything else in your org, governed by the running user's permissions and by whichever tools and context fields your administrator chooses to configure.
Two honest boundaries keep expectations straight:
- The agent doesn't do your accounting. There is no prebuilt accounting agent or connector; REDA AI supplies the platform, your administrator decides what an agent may touch. Debits, credits, tax, and posting logic remain REDA Financials' job.
- The guards always run. Every rule documented in this knowledge base — posting gates, period closures, permission checks — applies to agent-written records exactly as it does to a user typing. An agent can be told to draft a document; nothing reaches your ledger except through the same validations.
Getting started
Install from the AppExchange
REDA AI ships as its own managed package — search "REDA AI" on the Salesforce AppExchange. Sandbox first, as with any package: Installing REDA AI.
Assign access
Two primary permission sets — REDA AI Admin (configuration) and REDA AI User (chat) — plus dedicated sets for widget and portal channels: App Access & Permissions.
Build a first agent
Name, persona, channel, tools, context fields — point-and-click, with an AI-assisted builder that proposes a starting configuration: Creating an Agent.
Try it free
A free tier exists for evaluation — current terms and credits are on the free trial page and pricing page.
Where to go next
- Automating Accounting with Flow — the automation layer both products share.
- Invoice OCR — the AI already embedded in REDA Financials.
- REDA AI Knowledge Base — the platform's own documentation, end to end.
Frequently asked questions
No — it's a separate product with its own AppExchange listing, licensing, and credit plans (see pricing). What you gain by running both is architectural: one org, one security model, and agents that treat your accounting records as first-class Salesforce data.
Only what the logged-in user could see anyway, and only where the administrator has configured tools or context fields. Field-level security, sharing rules, and profiles are enforced before any record context loads — sensitive fields stay invisible unless explicitly configured and permitted.
An agent acts through the same door as everyone else: it needs a user with posting rights, an action tool that permits the write, and — in interactive channels — the user's confirmation on a card before the change executes. And every REDA Financials validation and posting guard still runs on the write.
Your records stay in your org. For each interaction, only the assembled context you configure — persona, instructions, configured context fields, the user's message — is sent to the AI provider for reasoning. REDA receives no copy; the provider does not use it to train models and deletes requests within 30 days by default under its published terms. The governance page documents the full data path.
No — those remain embedded features of the accounting product, tuned for their specific jobs. REDA AI adds the general-purpose layer: conversational agents, document intake, portal and website assistants, and Flow-driven AI automation across the whole org.
Usage is metered in credits — each interaction consumes a few, based on the text processed, from a monthly org-wide pool with per-user and per-channel caps available. Plans and current numbers live on the pricing page; there are no per-seat fees.
Was this article helpful?