AgentDx 1.0 · for Salesforce developers and admins
AgentDx writes your code automatically - just give it a prompt and let it cook!
Org grounding
Ask a general-purpose assistant to email an Account’s owner on create and it will happily write a new AccountTrigger, into an org that already has one. Two triggers on one object run in an order Salesforce doesn’t guarantee. AgentDx reads the org first, tells you when what you asked for already exists, and extends it instead of duplicating it.
You ask
“Send an email to the owner when an Account is created.”
Org preflight
Before it writes anything, AgentDx checks the org for automation that already runs when an Account is created.
ALREADY_EXISTS
One trigger and three active flows run after insert. AccountTrigger hands off to AccountTriggerHandler, and the active Account_Welcome_Email flow already sends an email alert.
It asks
A second trigger on Account would run in an undefined order, so it won’t write one. It asks whether to change what the existing flow sends, or whether this is a different email.
How it works
The same five steps for a question about the org and for a change to it.
Sign in through your browser. AgentDx uses the Salesforce CLI’s own connected app, so there is nothing to install in the org first.
It indexes the org in stages: the basics in seconds, objects and automation in a minute or two, code in the background. Ask questions meanwhile; answers come live from the org.
Ask about the org, or describe what to build. A preflight checks what already exists before anything is written.
Changes arrive as cards: Apex as a diff, Flows element by element, permission sets as a before-and-after matrix.
Every change is journaled before it runs. Revert from the Changes panel, with a plain list of what a revert can’t restore.
Features
Everything here ships in 1.0 and works against real orgs: production, sandboxes, developer and scratch orgs.
AgentDx indexes the org and checks it before every write, so it builds on what you have instead of beside it.
Changes arrive as cards you can actually read. Production always asks, and the riskiest changes make you type the org’s alias.
A journal records every change before it runs, and reverts what Salesforce lets it revert, in one click.
Approval cards
Before a change reaches the org you see exactly what it is. Apex arrives as a line diff, with the tests that will run and any lint finding beside the line it belongs to. A Flow change is broken down element by element.
Permission matrix
A permission set change shows how many active users hold the set, then a before-and-after matrix of system, object and field permissions. Anything that reaches beyond a single record, like View All Data, is called out at the top.
Change journal
Each change is written to the journal before it runs and settled after, so even a half-applied deploy can be explained. Revert puts back what it can, and tells you in plain words what it can’t.
Workbench
Open your project in a real editor, built on Monaco, with the agent beside it. The org browser lists every metadata type in the org and pulls any component down into the project when you open it.
No tool that can change the org runs without a preflight on record for what it touches. The gate runs before permissions are even checked, so it holds when tools are allowed to run without asking.
A metadata deploy to production becomes a validation with RunLocalTests. The org compiles it and runs your tests; nothing changes until you approve the quick deploy.
A local lint catches SOQL in loops, hardcoded Ids and missing sharing as the code is written. Code Analyzer findings at severity 1 or 2 block the deploy, and a turn can’t end with deployed Apex untested, failing or under 75% coverage.
Production is red, sandboxes amber, scratch and developer orgs green, wherever the org appears. An org AgentDx can’t identify is treated as production.
How the colours workAsk what changed recently and AgentDx reads the Setup Audit Trail, the only record of changes that leave no other trace, and answers from it.
Attach the requirements doc, a spreadsheet, a slide deck, a PDF or a screenshot. Word, Excel and PowerPoint files are read as text; images and PDFs go to the model as they are.
When AgentDx has to ask you something, each option comes with evidence from the org. It remembers your answer for that org, so a settled question stays settled.
Safety
Nobody should have to read a username to know they are about to change production. AgentDx colours every org by risk, everywhere it appears: the org list, the header and every approval card.
Production
Every change asks first. Deploys validate before they apply, and anything riskier than a small change needs the org’s alias typed in.
An org it can’t identify
Fails closed. It gets production-grade caution until AgentDx can tell what kind of org it is.
Sandbox
Small, reversible changes can run without a card. Larger ones, and anything it can’t classify, still ask.
Scratch and developer orgs
The place to try things, where a mistake is cheap. Still preflighted, gated and journaled.
Privacy
The org index, the change journal and your settings live on your Mac. Prompts pass through AgentDx’s servers on their way to Claude, and are never kept there.
Sign in with Apple, Google or email. We keep your email, your plan and what each agent request cost, never what it said.
Sign-in goes through the Salesforce CLI’s auth store, the same one sf uses. AgentDx keeps no passwords or tokens of its own.
The agent reaches Claude through AgentDx’s servers, which pass each request to Anthropic without storing or logging it.
AgentDx sends no analytics or crash reports, and turns telemetry off in everything it runs: the Claude agent, the Salesforce MCP server and the bundled Salesforce plugin.
Requirements
A Mac, an org and an AgentDx plan. That’s the list.
For the first 100 founding members, $99 a year with a 7-day free trial, or $9.99 a month. After the 100th founding seat, prices double to $199 a year and $19.99 a month. Founding members keep their price for as long as they stay subscribed. The pricing page shows how many founding seats are left.
No. Claude is included in every plan: AgentDx runs the agent through its own Anthropic account, so there is no key to create and no separate Anthropic bill. Each plan includes a monthly usage allowance, and the app shows how much of it is left.
Yes. Your plan belongs to your account, so sign in with the same Apple, Google or email account on any Mac you use.
AgentDx signs in through the Salesforce CLI’s own connected app, so most orgs need nothing installed first. But orgs can block connected apps that haven’t been installed in them: since September 2025, using an uninstalled connected app needs the Approve Uninstalled Connected Apps user permission, and orgs with API Access Control can restrict sign-in to allowlisted apps. If the Salesforce CLI connected app was never installed in your org and you don’t have that permission, Salesforce refuses the login. An admin can install the Salesforce CLI connected app in the org (Setup, Connected Apps OAuth Usage) or grant you the permission.
No. Sign-in happens in your browser, and the connection is saved in the Salesforce CLI’s auth store, the same one sf uses, so sf can use it too. AgentDx keeps no Salesforce credentials of its own.
No. In production every change asks first, metadata deploys run as a validation before they apply, and anything riskier than a small change needs you to type the org’s alias. Hard deletes are switched off in every org.
Most of them. Every change is journaled before it runs. Metadata deploys and record updates can be reverted from the Changes panel, and the revert tells you what it can’t restore, like LastModifiedDate. Anonymous Apex can’t be undone, and AgentDx says so before you run it.
Your prompts, your attachments and the org context the agent reads to answer them go to Anthropic’s API through AgentDx’s servers, which pass them on without storing or logging them. We keep your account email, your plan and a count of what each request cost, never its content. The app sends no telemetry. The privacy policy has the detail.
No. Node.js, a Java runtime for the Apex analyzer and the Salesforce tooling all ship inside the app. It runs on a Mac with nothing else installed.
macOS 13 Ventura or later, on Apple Silicon or Intel. There is one download for each; the download page picks the right one.
Claude Sonnet 5.5, through Anthropic’s Agent SDK. AgentDx decides how hard it thinks: lightly for questions about your org, and more deeply, up to its maximum, for changes, by how complex they are. Its Salesforce know-how comes from Salesforce’s own official Agent Skills and the Salesforce MCP server, both bundled with the app.
No. AgentDx is made by KineticBit Inc.. It is independent and not affiliated with or endorsed by Salesforce.
AgentDx for Mac. Start a 7-day free trial, connect an org, and ask it what already runs when an Account is created.
Download for Mac
