Go from plain-English request to deployed Salesforce delivery across fields, secure Apex, LWC bundles, permission sets, bounded updates to named existing Flows, list views, dashboards, and more. Clean runs move automatically; your team steps in only when a run needs it. Underneath it all, an intelligence layer scans your org, remembers what it learns, and watches for risky change.
Type the change you need. The same intake pass that drives the build locks onto the entities in your org — blue is resolved, amber will be created or asked about — and resolves to a single readiness verdict before anything runs.
blue = things the machine resolved in your org · amber = things it will create or ask a human about. Nothing else earns pixels.
SF Agent crawls your org into a queryable graph before it plans anything. Every proposed change is checked against what actually exists — and the blast radius is computed before a single line deploys. Try it: hover and click the graph.
Turn a plain-English request into a deployment-ready Salesforce change instead of bouncing between Setup, code, testing, and manual metadata work.
Stop hand-editing metadata and clicking through Setup. Describe the Salesforce change you need, let SF Agent generate the plan, and auto-deploy the clean runs.
Supported build work: custom fields, validation rules, flows, Apex classes, triggers, test classes, LWC bundles, permission sets, layouts, record types, list views, reports, dashboards, custom settings, report types, and more.
Human in the loop when it matters: if the AI runs into ambiguity or a blocker, users can provide input directly on the run so SF Agent can continue with the right context.
Transform Workbench Beta remains available as a secondary lane for modernization work.
Try It FreeSF Agent means something specific to each person who touches Salesforce change — from the boardroom to the debugger.
Turn a backlog that took a consulting statement-of-work into governed, plain-English changes. Every run leaves a tamper-evident audit trail, and the platform is sandbox-first — it never touches production uninvited.
Stop clicking through Setup and hand-editing metadata. Type what you need in plain English; SF Agent plans it, validates it against your real org, and deploys the clean runs — asking you only when a decision is genuinely yours to make.
Bulk-safe, FLS-aware Apex with genuine test coverage, LWC bundles, and triggers — compiled into one atomic Metadata API package, check-only validated, and identity-hashed so what you approve is exactly what deploys. Or download the package for your own CI/CD.
Human approval gates on the exact compiled artifact, a source→plan→package→approval→deploy evidence chain, least-privilege permission sets, AES-256 encryption, and SSO with policy-based MFA. The platform validates against a sandbox, so prod stays yours.
Generate supported Salesforce primitives from natural language and ship clean runs into Salesforce, with a human step only when the run needs it.
Custom objects, fields, record types, validation rules, page layouts, and picklist values. Generated from your description, deployed via Metadata API.
Flows, Apex classes, Apex triggers with test classes, LWC bundles, approval processes, assignment rules, list views, dashboards, custom settings, and report types. Each operation goes through preflight validation before reaching your org.
Clean runs can move straight through to deployment. When a change needs confirmation, SF Agent surfaces the plan so you can inspect exactly what will change before approving.
Modernization is a secondary lane when you need it: break Profiles into Permission Sets, reconstruct Workflow Rule or Process Builder behavior, and implement the best-fit secure Apex or non-Flow design.
Three-layer safety: preflight checks before deploy, quality gates on the spec, and post-deploy verification that fields exist, flows activate, and Apex compiles.
Deploy failures are classified into structured fix options, including retry, rename, or skip paths, instead of raw Salesforce error dumps and manual guesswork.
These are the Salesforce primitives SF Agent is positioned to generate, validate, and deploy from plain-English requests, with human review only when the run needs it.
These primitives are being built and tested. They will move into supported status as each one passes live deployment validation.
Security controls are part of the architecture and remain subject to customer validation, deployment configuration, and production assurance.
Decide how your team deploys. Clean runs can move automatically, higher-risk changes can require a human step, and every run stays visible with traceable audit history.
All secrets and API keys encrypted at rest with AES-256-GCM. Random IV per encryption. Your BYOK keys never stored in plaintext.
Use Salesforce SSO via PKCE OAuth, or email and password with email or TOTP challenges. Tenant policy, enrollment state, trusted devices, and email-service configuration determine when MFA is enforced.
Use your own Anthropic, OpenAI, or Azure key. Full control over model selection and cost. Keys encrypted at rest, used only in-flight.
SHA-256 hash-chained audit log. Every deployment, login, and admin action recorded with tamper-detection built in.
Four roles: Admin, Developer, QA, Viewer. Tenant-scoped server-side authorization is designed into protected operations and is continuously covered by security and negative-authorization testing.
Three policy levels: block all model calls, allow summaries only, or require approval. Control exactly what data leaves your tenant.
HttpOnly secure cookies, SameSite=Lax session handling for OAuth compatibility, SameSite=Strict trusted-device cookies, origin validation on mutating APIs, and server-side rate limiting for auth and run creation.
SF Agent connects delivery work to governed org context: retained scan evidence, permission-aware memory, drift analysis, recommendations, and decision records.
Completed scans, builds, and governed drift events can contribute durable, source-linked memory. Ask questions in plain English and inspect the org evidence used in each answer.
When scheduled monitoring is configured, Change Guardian compares scans, detects supported drift, estimates blast radius, ranks risk, and creates evidence your admins can review.
A diagnostic mapped to Salesforce Well-Architected concerns such as legacy automation, security posture, technical debt, and API-version age. Supported findings include evidence, impact context, and priority.
Planning AI agents on your org? Use a readiness assessment across observable data configuration, automation health, security, and metadata hygiene, with cited gaps for human review.
Covered AI and human decisions are written to hash-chained records so later mutation or missing links can be detected. Governance exports preserve the evidence for independent review.
When validation fails, bounded repair attempts can classify Salesforce errors, propose package edits, and revalidate. The run stops for review when policy, confidence, or retry limits require it.
Org snapshots serialize into clean SFDX source format — the same diff-friendly structure Salesforce tooling uses — for a version history you own. Your org’s state stays reviewable, comparable, and yours.
Eligible failures can become structured lessons. Recurring patterns are proposed as governed rules and must pass review and evaluation gates before influencing future runs.
Tracked model calls expose usage and estimated cost by run, purpose, and model. Bring your own supported provider key and use routing policies and budgets to control spend.
Round nodes are machine stations that run without you. The amber diamond is the only place a human is interrupted — and only when the answer is genuinely unknowable. Everything with a safe default gets the default; everything the AI breaks, the AI repairs first.
THE RULE: a human is interrupted only when the answer can’t be inferred safely. Clean runs move automatically, end to end.
What the model was asked, what it answered, what the org validated, who approved what — a tamper-evident trail on every run. This is what no consulting invoice ever includes.
A $200K consulting engagement ends with a slide deck. A run ends with receipts.
SF Agent assembles inspectable packages and Salesforce validation evidence. Your release policy and authorized reviewers decide whether a production deployment proceeds.
Supported changes are compiled into a deployment package — package.xml plus metadata — that you can download and inspect before deployment.
Check-only validation runs against the selected target org, with Apex test level selected by environment and policy. The resulting Salesforce response becomes part of the run evidence.
If validation fails, the Repair Agent can classify the returned errors, propose an edit, and revalidate within explicit retry, confidence, and approval limits.
When Salesforce reports the validation as eligible, an authorized user or policy can request quick deploy using its validation reference. The outcome is recorded in the audit and governance evidence.
Three steps from requirement to deployed Salesforce work — watch it happen end-to-end.
Sign up and connect through a Salesforce Connected App using guided OAuth setup. No managed package or custom in-org component is required for the current integration model.
Type your requirement in plain English. SF Agent generates the plan, runs validations, and prepares the build path for deployment.
Clean runs deploy automatically. If the run needs confirmation, you step in, approve the change, and the platform completes deployment with verification.
SF Agent is in evaluation — use the available workflow in a sandbox, review its evidence, and help shape the production offering. Paid plans come later; early design partners get first access.
Explore the currently available build, deployment, scan, monitoring, and governance capabilities in a sandbox evaluation.
During evaluation the platform is free — model usage is billed by your own AI provider. Pricing plans will be announced later.
Short answers to the questions serious buyers, admins, and platform teams ask before they trust automation in a Salesforce org.
Connect your Salesforce org, describe one real request, and see how SF Agent handles the build path from natural language to deployed Salesforce work.
Start Free Trial