See the Risk.
Clear the Debt.
Protect What’s Next.

When Horus Eye’s AI reads your org, it reads metadata — never records. Context is allowlisted, classified and sanitized before it leaves; one gateway makes every call; an output firewall checks every answer; and every request carries an audit id and an AI Data Scope manifest you can open. AI is off by default, per org, with a written record of every change.

See the Trust Layer in full

The job usually arrives as a question someone senior asks in a meeting, and nobody in the room can answer with evidence:

Different questions, one root: nobody holds an evidence-backed picture of the org as a system. The people who built it left. The documentation describes 2019. Every new consultant spends their first month rediscovering it from scratch — and their findings leave with them, too.

Nobody sets out to build technical debt; it accretes from perfectly reasonable decisions. Each admin adds automation in the technology of their era — Workflow Rules, then Process Builder, then Flow — and retirements never quite finish, so three generations run side by side on the same object. Under deadline it is always safer to clone than to modify, so the org accumulates near-copies that drift apart. Permission sets get created per request and never withdrawn, because withdrawing access is how you break someone’s Monday. Managed packages weave their namespaces into custom code. And each of these choices is invisible at the moment it is made — it only becomes visible years later, as the aggregate: an org where every change touches more than anyone expects.

That is why counting things doesn’t help. A metadata count is not an insight — the count was never the problem:

Horus Eye is built to produce the second kind of sentence, about your org, with the evidence attached: the automation, dependencies and architecture patterns that make your Salesforce org harder to change than it needs to be — found, named, and where safe, cleared.

Think of it as a Salesforce org health check with an architect’s depth: Horus Eye connects read-only, inventories the org, and runs 90 detection rules across seven health dimensions. The result is an Estate Health score (0–100) — severity-weighted exposure measured against the components examined: a density of evidence, not a grade of how the org is run — with a per-dimension status and a scan-over-scan trend.

The seventh dimension, Reporting & Analytics, currently reports “no checks yet” — because a dimension nothing looked at must never read as clean. That coverage honesty runs through the product: every rule reports what it examined, what was clean, and what it could not read.

Every finding carries one or more business-impact lenses — Financial & Cost, Security & Reputation, Compliance & Governance, Scalability & Sustainability, Operational Reliability, Data Quality & Decision Integrity, Modernization & AI Readiness — so an executive can read the same queue an architect works from.

Findings are worded the way an architect would word them: they “appear”, they “may”. Nothing in Horus Eye is a verdict, and nothing on this page is either — the evidence is the product.

Select any Flow, object, field, permission set, Apex class, report — most metadata types — and ask for a Deep Read. Horus investigates against the org’s own structure using read-only retrieval tools, then explains the component’s apparent business purpose, what depends on it, what it overlaps with, whether it is healthy, and what changing it could affect — with dependency, save-path, permission-path and blast-radius diagrams when the context calls for them.

Horus Eye distinguishes directly observed configuration from AI-inferred business intent and shows the evidence behind its conclusions. Every Deep Read is model-stamped, dated and confidence-rated; AI is off by default and per-org opt-in.

Deep Read has its own page — including exactly what is sent to the model (structured scan evidence, never record data) and the full safety model.

Assessment tools stop at the list. Horus Eye’s Home says something different: on the demo org, 35 of 1,577 findings can be cleared now — and “Clear the safe ones” is a button, not a recommendation. Here is what pressing it actually means:

1. Horus proposes a fix plan. Numbered steps, each one of sixteen sanctioned, reversible operations — tightening access, retiring unused permission sets, cleaning inactive-user assignments, consolidating record types and validation rules, decomposing a god object into extension objects. Anything outside that vocabulary stays a guided Setup step.
2. You approve each step — or don’t. The one question only the owner can answer is asked before anything else.
3. Validation runs with zero writes. Only after it passes does Fix it execute.
4. Everything is reversible. Rollback re-grants exactly what was removed. Managed-package components are refused outright.
5. The next scan proves it. Findings clear themselves when the evidence is gone — the re-check distinguishes “resolved” from “still present, but changed”.

Prefer your own pipeline? Export the planned change as a Metadata API or SFDX bundle and promote it through your CI/CD instead. And what no tool should touch alone — Flow consolidation, domain redesign, the judgment calls — Kemisoft’s architects clear with you. Same promise, human half.

Either way, Horus Eye stores counts, flags, names and derived metadata only — never raw record data, never Apex or Flow source bodies, never template contents or query text. Scans never change the org; writes need the deployment switch, a per-org grant, and your per-step approval, all at once.

The report card ships with a coverage tab (what was examined, what could not be read), the business-impact lenses, a Now / Next / Later roadmap, and a branded PDF export. The Impact pack turns any selection of findings into a stakeholder-ready PDF or email. More on the report card.

Does Horus Eye change my org?

Not unless you tell it to — three times over. Scans are read-only. Writes require the deployment switch, a per-org write grant in Settings, and your per-step approval; every executable step is validated with zero writes first, is reversible, and is verified by the next scan.

What does Horus Eye store?

Counts, flags, names and derived metadata. Never raw record data, never metadata bodies (Apex or Flow source), never template contents or query text. Deep Read sends structured scan evidence to the model — never record data.

How is this different from Salesforce Optimizer or a health-check checklist?

Optimizer and checklists report configuration statistics. Horus Eye connects metadata, dependencies, business logic, permissions and automation into findings with evidence, explains the business impact through seven lenses, proposes the fix — and applies the safe subset itself, reversibly. A count tells you the org is big; a finding tells you what to do on Monday.

How long does a scan take?

About one to two minutes end to end on a roughly 3,200-component demo org. Bigger orgs take longer; every scan reports exactly what it examined and what it skipped.

Which orgs and editions?

Production, sandbox and scratch orgs, with the login host chosen per org (including My Domain). If a capability isn’t available in your edition, its checks report as skipped rather than silently passing.

Is Horus Eye a Salesforce org health check?

It starts as one — a read-only Salesforce org health check producing a scored, evidence-backed picture in minutes — and then goes where checklists stop: dependencies, blast radius, AI-assisted Deep Reads, and reversible fixes for the safe findings.

Do I need to install a package?

No. Quick Connect installs nothing and stores nothing; OAuth mode uses a one-time External Client App in your org.