# Specialist AI agents for Salesforce, with limits you set.

> SyncOnAI 360 agents are specialist versions of the org-aware chat, each with its own instructions, skills, tools, knowledge and model. The library covers Salesforce admin, Flow and automation, Apex, Lightning Web Components, and SOQL and data work. Agents run on the same chat engine with the same approvals, and every tool call is recorded.

Source: https://synconai360.com/features/agents

## Key facts

- **5**: Specialist agents in the library
- **3**: Per-tool settings: allow, deny or confirm
- **Every**: Tool call recorded
- **0**: Memories saved without a person's approval

## Why a single general AI assistant struggles with Salesforce work

Salesforce work spans very different jobs: configuring objects, building Flows, writing Apex, building components, answering data questions. One general assistant with every tool and no limits is both less useful and less safe.

- **Jack of all trades.** A general assistant gives general answers. Specialist work needs specialist instructions and tools.
- **Too much power by default.** An assistant that can do everything can also do the wrong thing, with nothing in between to stop it.
- **No visibility.** Teams cannot see which tools an assistant has, or whether a tool writes or only reads.
- **Memory nobody approved.** Assistants that remember things on their own pick up wrong conventions and never let go.

## Which Salesforce AI agents are available?

The Salesforce Admin agent works on objects, fields, validation rules, layouts, list views and reports. The Flow and Automation agent builds and fixes Flows. The Apex Engineer writes Apex classes, triggers and tests. The LWC Developer builds Lightning Web Components. The SOQL and Data agent answers read-only queries and data questions.

Each agent is built on Salesforce's own published skills, adapted to the tools in SyncOnAI 360, and works from the connected org's metadata and source rather than from a generic picture of Salesforce.

- Salesforce Admin
- Flow and Automation
- Apex Engineer
- LWC Developer
- SOQL and Data

## How do you control what a Salesforce AI agent can do?

On the Agents page, each agent shows its tools and whether each one reads or writes. Any tool can be allowed, denied or held for your confirmation, so an agent's reach is a decision rather than a default.

While an agent is active, any change to an org waits for approval and becomes a proposal like any other change, with named checks, blast radius and, for production, a second admin's approval. Every tool call is recorded, and recent runs and their outcomes are listed on the Agents page.

- Tools listed with read or write
- Allow, deny or confirm per tool
- Changes still go through proposals
- Every call recorded

## How do agent skills and memory work?

Skills are reusable instructions an agent can load when it needs them. Each skill records where it came from and which version is in use, so behaviour can be traced.

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, never cross into another workspace, and can be reviewed in settings.

- Versioned, traceable skills
- Memories only with approval
- Memories never leave the workspace

## Which models do Salesforce agents run on?

The workspace chooses a model for agents separately from the model the chat uses, from Anthropic and OpenAI. When AI is included in the plan, the right model is picked per task: a light model for quick answers, a stronger one for building and the strongest for hard reviews.

Neither provider trains on API traffic. On the Free plan, agents run on your own provider key and account.

- Separate model for agents and chat
- Model picked per task when AI is included
- No training on API traffic

## How are agents different from the chat?

The chat is the general assistant. Agents are specialists with their own instructions, skills and tool permissions for one kind of work, such as Flows or Apex. Choose an agent when the job is clearly in its area, and the chat when a question spans several.

Both run on the same engine, read the same org and follow the same approvals, so switching between them never changes what is allowed to reach the org.

- Specialists for one kind of work
- One engine and one set of approvals

## Working with an agent

1. **Pick.** Choose the specialist in the chat composer.
2. **Ask.** Describe the job; the agent works from the org's metadata and source.
3. **Confirm.** Tools held for confirmation wait for you.
4. **Review.** Changes arrive as proposals with checks and blast radius.
5. **Approve.** Production changes need a second admin; every call is on record.

## AI assistance, general versus specialist

| Without | With SyncOnAI 360 |
|---|---|
| One assistant for every job | Specialists for admin, Flow, Apex, LWC and data |
| Every tool on by default | Allow, deny or confirm each tool |
| Unknown read or write power | Each tool labelled read or write |
| Memory that builds up unchecked | Memories saved only with approval |
| No record of what it did | Every tool call and run recorded |

## Who works with the agents

- **Salesforce admins.** Hand routine configuration to the Salesforce Admin agent, objects, fields, validation rules, layouts, list views and reports, and review each change as a proposal before anything reaches the org.
- **Salesforce developers.** Pair with the Apex Engineer and the LWC Developer for classes, triggers, tests and components that follow the org's own patterns, with every tool call recorded.
- **Platform owners.** Decide exactly what each agent may do: allow, deny or hold any tool for confirmation, see recent runs, and keep production behind a second admin's approval whatever an agent proposes.
- **Consultancies.** Give every consultant the same specialist agents and approved memories, such as a client's naming conventions, so work stays consistent across the team without crossing between workspaces.

## Frequently asked questions

### What is a Salesforce AI agent in SyncOnAI 360?

A specialist version of the org-aware chat with its own instructions, skills, tools, knowledge and model, for jobs like Flow building, Apex or data questions.

### Can agents deploy changes on their own?

No. Any change an agent makes to an org waits for approval and goes through a proposal with checks; production needs a second admin.

### Can we restrict an agent's tools?

Yes. Each tool can be allowed, denied or held for confirmation.

### Is every agent action logged?

Yes. Every tool call is recorded, and recent runs and their outcomes are listed on the Agents page.

### Do agents remember our conventions?

Only what a person approves. Agents can suggest memories; nothing is saved until approved.

### Is the SOQL and Data agent able to change data?

It is for read-only queries and data questions. Data-changing Apex in the chat always waits for an explicit Allow.

### Can chat and agents use different models?

Yes. The workspace sets a model for chat and a separate one for agents.

### Are agents included on every plan?

Yes. Every plan includes every feature.