# An AI agent for Salesforce that actually knows your org.

> SyncOnAI 360's AI agent is a chat assistant that works from a copy of your Salesforce org's metadata and source. It picks the right tool for each question, shows which tools ran and cites its sources, and builds changes as validated proposals that wait for a person to approve them.

Source: https://synconai360.com/features/ai-chat

## Key facts

- **5**: Specialist agents: Admin, Flow, Apex, LWC and data
- **0**: Changes deployed without a person approving them
- **2**: Model providers: Anthropic and OpenAI
- **Every**: Reply shows which tools ran and cites its sources

## Why general AI assistants struggle with Salesforce

General-purpose assistants know Salesforce in the abstract. They have never seen your org, so they invent field names, rebuild automation you already have, and hand you code with no way to check it against the org before it ships.

- **They guess your org.** Without your objects, fields and Flows in front of it, an assistant writes for a demo org. Every answer needs checking against the real thing.
- **They duplicate what exists.** Ask for a Flow to route cases and you get a new one, even though two Flows already run on that trigger. The org gets more complex with every request.
- **No line to production.** Code pasted from a chat window has no validation, no approval and no record. The speed it adds is lost again in review, or worse, in incidents.
- **No accountability.** When an assistant gets something wrong, there is no record of what it looked at, which tools it used or who accepted its output.

## What can you ask a Salesforce AI agent about your org?

Ask in plain language and the agent picks the right tool for the question: reading metadata, following the dependency map, running a live SOQL query, checking the health score, pulling audit findings or searching the Knowledge Hub. Every reply shows which tools ran, and sources from the Knowledge Hub or the web appear as numbered citations under the answer.

Questions like "What writes to Opportunity.Amount?", "Which Flows on Case overlap?" or "Is any of this Apex unused?" get answers from your org, not a generic one. Under any answer you can ask for it simpler, more technical, as a table or as a recap, or save it as a page.

- Answers from your org's metadata, dependencies, health and audit
- Live SOQL against the org when the question needs data
- Tools shown and sources cited on every reply
- Reshape any answer, or save it as a page

## How does the AI agent build changes safely?

Describe what you want built. The agent checks what already exists in the org first, drafts the change to match how your org is built, validates it against the org and saves it as a proposal. Nothing is deployed until someone approves it, and production needs an admin other than the person who asked.

Plan Mode lets the agent think a change through without being able to make it. Anonymous Apex that inserts, updates or deletes data waits for you to press Allow before it runs. Name a Flow, page or component and it opens in a visual builder beside the chat, so you can review or adjust the change by hand.

- Checks for existing solutions before building
- Every change validated and saved as a proposal
- Plan Mode for thinking without the ability to change
- Data-changing Apex waits for an explicit Allow

## What are specialist Salesforce agents?

An agent is the chat with a job: its own instructions, skills, tools, knowledge and model, all running on the same engine with the same approvals. The library includes a Salesforce Admin agent for objects, fields, validation rules, layouts and reports; a Flow and Automation agent; an Apex Engineer; an LWC Developer; and a SOQL and Data agent for read-only questions.

Each agent lists its tools and whether each one reads or writes, and each tool can be allowed, denied or held for confirmation. While an agent is active, any change to an org waits for approval, and every tool call is recorded.

- Admin, Flow, Apex, LWC and data specialists
- Built on Salesforce's own published skills
- Per-tool allow, deny or confirm
- Every tool call recorded

## How do you control what the AI remembers and which model it uses?

Agents can suggest something worth remembering, such as a naming convention your team uses, but nothing is remembered until a person approves it. Memories belong to a conversation, a project or a person, and never cross into another workspace.

The workspace chooses its models from Anthropic and OpenAI, which do not train on API traffic. On the Free plan every AI call runs on your own provider key and account; paid plans include AI, and you can still bring your own key if you prefer.

- Memory only with a person's approval
- Models from Anthropic and OpenAI, no training on API traffic
- Your own key, or AI included in a paid plan

## How the AI agent works

1. **Ask.** Type a question or a request, or pick a specialist agent for the job.
2. **Investigate.** The agent reads your org's metadata, dependencies, health and audit, and shows its tools.
3. **Draft.** For a change, it checks what exists, drafts to your conventions and validates against the org.
4. **Review.** Open the change in a builder, adjust it, and send the proposal for approval.
5. **Approve and ship.** A second admin approves production, and the deploy leaves a receipt you can roll back from.

## AI for Salesforce, general assistant versus SyncOnAI 360

| Without | With SyncOnAI 360 |
|---|---|
| Answers written for a demo org | Answers from your org's own metadata and source |
| New automation on top of old | Checks what already exists before building |
| Code pasted into the org by hand | Validated proposals with checks and approval |
| No record of what the AI did | Every tool call recorded, every deploy receipted |
| One general model for everything | Specialist agents with their own tools and permissions |

## Who uses the AI chat

- **Salesforce admins.** Ask questions about the org in plain language and have configuration drafted as a proposal you review, rather than clicking through Setup.
- **Salesforce developers.** Get Apex, Flows and components drafted against the org's real metadata and patterns, with every tool the AI used shown under the answer.
- **Consultancies.** Give every consultant an assistant that already knows each client's org, so the first day on a new account is productive.
- **Platform owners.** Use AI on production orgs with every change held as a proposal, checked, validated and approved by a second admin.

## Frequently asked questions

### What is a Salesforce AI agent?

A Salesforce AI agent is an assistant that can answer questions about an org and build changes to it. SyncOnAI 360's agent works from your org's own metadata and source, shows its tools and sources, and turns every change into a validated proposal.

### Can the AI agent change our org without approval?

No. It can read, analyse and draft. Every change becomes a proposal with named pre-flight checks, and nothing is deployed until a person approves it. Data-changing Apex run from the chat waits for you to press Allow.

### How is this different from Agentforce?

Agentforce runs AI agents inside your Salesforce org for your end users. SyncOnAI 360 is the AI that helps your team build and configure Salesforce itself, including your Agentforce agents.

### Which AI models does it use?

Models from Anthropic and OpenAI, chosen by the workspace. Neither provider trains on API traffic. On Free you use your own key; paid plans include AI.

### Does the agent see our customer records?

It works from your org's metadata, which SyncOnAI 360 stores. When a question needs data, it can run a live SOQL query against the org, and the records are not stored by SyncOnAI 360.

### What does Plan Mode do?

Plan Mode lets the agent think through and plan a change without being able to make one. It is useful for scoping work, reviewing options or explaining an approach before anything is drafted.

### Can we limit what an agent is allowed to do?

Yes. Each agent lists its tools and whether each reads or writes, and each tool can be allowed, denied or held for confirmation. Every tool call is recorded.

### Does the agent remember things between conversations?

Only what a person approves. Agents can suggest a memory, such as a naming convention, and it is saved only after approval. Memories stay within your workspace.

### Can it work with files and images?

Yes. Attach files to a message and they stay with the conversation. The agent can produce files, such as a PDF write-up, and images when the workspace has an OpenAI key.

### What is the prompt library?

A verified library of prompts opened from the Prompts button in the chat. Pick one, fill in its fields with pickers that read your org, and it runs in the chat. Prompts also have their own page with history.

### Can we organise and search past conversations?

Yes. Search every chat from the sidebar, rename chats and group them into projects. A temporary chat leaves nothing behind when you close it.

### Is the AI agent included on the free plan?

Every plan includes every feature. Free covers one user and one production org with its sandboxes, on your own Anthropic or OpenAI key, with no credit card and no time limit. Paid plans include AI and cover more people and production orgs.