Security

VanguardOps is built by an architect who’s audited 80+ enterprise Salesforce orgs. The security model isn’t a marketing afterthought — it’s the entire reason the product exists in the form it does.

This page explains, in plain language, what VanguardOps accesses in your Salesforce org, what it never accesses, and exactly what happens when a fix is executed.

What we access

VanguardOps reads Salesforce metadata only via OAuth-scoped API access:

• Custom and standard objects, fields, and relationships
• Flows, validation rules, and workflow rules
• Permission sets, how many users hold each one, and license counts
• Apex classes, triggers, and test coverage data
• Connected applications, named credentials, certificates, and auth providers
• Org configuration for Sales Cloud and Service Cloud
• Org limits, and an approximate record count for each custom object. This is a number Salesforce reports, used to judge whether an object is in use. No record is opened to get it.

These are standard Salesforce APIs: the Tooling API, object describe calls and setup-object queries, the same ones the Salesforce CLI uses. The scans run from outside your org. To let your org sign in, your admin installs one small managed package, VanguardOps Connect, once. It holds only the VanguardOps External Client App and its OAuth settings: no Apex, objects, fields or flows. You can uninstall it from Installed Packages at any time.

What we never access

VanguardOps never queries:

• Customer records of any kind — Contacts, Accounts, Leads, Opportunities, Cases, custom object data
• Reports, dashboards, or report data
• Email, calendar, task, or activity history
• Files, attachments, or Salesforce Files content
• Chatter posts or feed data

To be precise about how: VanguardOps requests Salesforce’s api and refresh_token scopes, and the api scope on its own could read records. So the safeguard is in our code, not the scope: the scanners only call metadata APIs, and automated tests check that they stay that way.

Authentication and token storage

VanguardOps uses OAuth 2.0 with PKCE (Proof Key for Code Exchange). When you connect your org:

1. You authenticate directly with Salesforce. VanguardOps never sees your Salesforce password.
2. Salesforce issues a refresh token for the two scopes VanguardOps requests: api and refresh_token.
3. The refresh token is encrypted at rest before being stored in our database, with the encryption key held separately from the database and managed via secrets management.
4. Access tokens are short-lived. VanguardOps uses them only when someone on your account runs a scan, asks the AI Architect about the org, or applies a change they approved in a sandbox.

You can revoke VanguardOps access at any time from your Salesforce Setup → Connected Apps OAuth Usage page. Revocation is immediate and irreversible — VanguardOps cannot re-establish access without you completing a new OAuth flow.

On-demand scans, no persistent agents

Scans are on-demand only. VanguardOps does not run continuous monitoring, scheduled scans, background polling, or persistent agents against your org. Every scan is initiated by a logged-in user, runs against the metadata APIs with your scoped OAuth token, and releases the connection when it completes. Between scans, your access token sits idle.

Scans only read. Nothing in your org changes during a scan.

How remediation is kept safe

Finding problems is read-only. Fixing them is where the safety mechanism matters, so here it is, stated exactly:

• Every fix requires explicit human approval. There is no autonomous remediation and no scheduled execution.
• Execution is sandbox-gated and refuses production by design. The execution path will not run against a production org.
• Six kinds of change are live in sandboxes today. VanguardOps can add fields, validation rules, flows, permission sets, Apex classes and Apex triggers. Before every change, it confirms with Salesforce that the org is a sandbox.
• Salesforce checks each change before you approve it. You see the exact before and after, Salesforce validates it, and nothing is applied until you approve that exact change.
• Every change that runs is logged and can be put back. Putting a change back removes what it added. Removing a new field also removes any values entered in it, so test in your sandbox before you repeat a change anywhere else.

Where your data lives

VanguardOps never moves your customer data. Customer records, contacts, opportunities, and all record-level data remain in your Salesforce org. Scan results include the Salesforce user IDs of whoever last changed a component. Scan results and remediation audit logs are stored in our database, hosted by Railway in the United States.

Deleting your data

You don’t have to email anyone. In the app, go to Settings, pick the org, choose Delete data and type DELETE. That permanently removes:

• The org connection and its stored tokens
• Every scan of that org, with its objects, fields, rules and automations
• Fix plans, recommendations and AI conversations about the org
• Audit log entries about the org

Three things are kept, on purpose:

• The first 15 characters of the Salesforce org ID and the account that claimed it, so a deleted org can’t claim a second free scan
• Benchmark figures keyed by a salted hash of the org ID, with no names and no raw IDs
• One log entry recording that the deletion happened

Disconnecting an org is the lighter option: it revokes the token at Salesforce and keeps your scan history.

Check it yourself

You shouldn’t have to take this page on trust. Three things you can verify in your own org:

• The consent screen. When you connect, Salesforce shows exactly what VanguardOps is asking for before you approve it.
• Setup → Connected Apps OAuth Usage. VanguardOps appears there once connected, with who authorized it. Revoke it there and access ends immediately.
• Your API logs. If your org has Event Monitoring, the API event logs record the requests VanguardOps made during a scan, so your security team can see for themselves which objects were queried.

Questions a security team asks

Is anything installed in our org?

Only the sign-in app. Salesforce lets an outside app sign in to your org only after it is installed from a package, so your admin installs VanguardOps Connect once. It contains the VanguardOps External Client App and its OAuth settings and nothing else: no Apex, no custom objects, no flows.

Why the api scope, if it only reads metadata?

Salesforce has no narrower scope that covers the Tooling API and setup objects a scan needs. The api scope could read records, so the limit is enforced in VanguardOps code, and automated tests check that the scanners don’t query record data.

Which AI sees our metadata?

Anthropic’s Claude, through its API. It receives scan metadata only, and email addresses found in that metadata are redacted before it is sent. Anthropic does not train its models on API customer data.

Where do the dollar figures come from?

A fixed formula, not the AI: hours to fix, a US labor rate for the role, and a yearly carrying cost by type of problem. The AI writes findings and sets the health score. The figures are directional estimates, not a quote.

Can it change our production org?

No. Scans only read, and the fix path refuses any org Salesforce reports as production.

Do you have a SOC 2 report?

Not today. VanguardOps is a small company and says so. What you get instead is on this page: exactly what is read, where it is stored, how to verify it and how to delete it.

Vulnerability reporting

If you discover a security vulnerability in VanguardOps, please report it to chris@vanguardops.ai. We will reply, and keep you updated until it’s fixed.

Please do not publicly disclose vulnerabilities before we have had a reasonable opportunity to address them.

Compliance

VanguardOps does not have a SOC 2 report today. For a security questionnaire, a data processing agreement, or questions about HIPAA, FINRA, ISO 27001 or GDPR, email chris@vanguardops.ai.

Questions

Architects asking pointed security questions get fast direct answers. Email chris@vanguardops.ai. If a question deserves a public answer, we add it here.

Last updated: November 11, 2026.