Features · Intelligence
Intelligence that shows its work
SOC is the layer that reads, enriches, and explains — never the layer that invents the numbers. It turns paperwork into records, datasheets into specs, and questions into grounded answers, while the deterministic engines keep ownership of every figure.
Documents in, data out
Reads the paperwork so nobody re-keys it
Order confirmations, quotes, emails, and PDFs are parsed into structured records — orders with lines, quantities, and tracking, not text in a notes field. A deterministic, rule-based extraction pass backs up the parsing, and imports land as reviewable records rather than silent writes.
- Order confirmations, emails, and PDFs parsed into structured orders
- A deterministic extraction pass backs up the parsing — two readers, one record
- Imports are reviewable before they become truth
The import path
- In
- PDFs, emails, confirmations
- Parsed
- structured lines and quantities
- Backstop
- deterministic extraction pass
- Out
- reviewable records
Product enrichment
A product library fed by real datasheets
SOC enriches products in your library with specifications drawn from manufacturer documentation — and keeps the provenance, so a spec traces to where it came from. Those specs matter more here than in a catalog: they are the same inputs the physics engines compute from.
- Specifications enriched from manufacturer documentation into your library
- Provenance kept — a spec traces to its source
- The enriched specs feed the physics engines: lens geometry, transmit power, patterns
Why specs matter here
- Enriched
- from manufacturer documentation
- Traced
- provenance on the spec
- Consumed
- by the physics engines
- Guessed
- never
The grounded assistant
Answers from your data, with receipts
The in-app assistant answers from your own operational data — projects, orders, renewals, products — with provenance on what it says. Numeric-claim guardrails hold its figures to records; when the data cannot support an answer, it says so instead of improvising.
- Grounded in your operational data, with provenance on answers
- Numeric-claim guardrails — figures trace to records, or SOC declines to state them
- Sensitive fields are redacted before any request leaves the platform, and restored on return
- Per-tenant limits and a usage audit trail your admins can see
Trust mechanics
- Grounding
- your operational data
- Figures
- traced to records, or declined
- Redaction
- sensitive fields masked in transit
- Audit
- per-tenant limits and logging
The trust invariant
SOC never invents the numbers — the physics and cost engines own every figure; SOC researches, suggests, and explains.
That is an architectural boundary, not a policy. The generative layer has no path to author a coverage value, a measurement, or a price.
Direction
Where SOC is going
Framed honestly: this is roadmap, not product. Everything above this section ships today.
The proposed next layer is SOC learning your org’s design conventions — the mounting heights you standardize on, the brands you prefer, the way your senior designers lay out a floor — from the projects you approve, with the same provenance discipline as everything above. It is a proposal on our roadmap, not a shipped capability, and we will move it onto the delivered list only when it is real.
The mechanics behind the trust
Structural facts about how SOC operates — no invented benchmarks.
- 0
- physics numbers generated by SOC
- the deterministic engines own every figure
- Provenance
- answers cite where they came from
- grounded in your operational data
- Redacted
- sensitive fields masked before requests leave
- restored only on return
- Per-tenant
- limits, logging, and an audit trail
- usage visible to your admins
Intelligence questions
What is SOC, exactly?
SOC is the intelligence layer of SiteOps Command. It reads documents and orders into structured data, enriches product specifications with provenance, and answers questions from your own operational data. It does not generate physics or cost figures — the deterministic engines own those.
Does SOC compute my coverage or cost numbers?
No — and that is the design. The physics and cost engines own every figure; SOC researches, suggests, and explains. When SOC cites a number, it comes from a record or an engine, not from generation.
What leaves the platform when SOC makes a request?
Requests pass through a safety pipeline that redacts personally identifiable fields before anything leaves and restores them on return. Usage is logged per tenant, with limits your admins can see.
Will SOC learn our design standards?
That is the direction, and we frame it honestly: an org-conventions learning layer is proposed on our roadmap, not shipped today. What ships today is what this page describes — parsing, enrichment, and grounded answers with provenance.
See grounded intelligence on your own paperwork
Bring a stack of order PDFs and a product list — talk to us about what SOC turns them into.