# Sightwell technical overview > Sightwell connects the industrial systems you already own and makes their operational knowledge safely usable by AI. Deployed in your Azure tenant or on Kubernetes you control. This page restates the public product description on [sightwell.ai](https://sightwell.ai/). It does not add deployments, certifications or connectors beyond that page. Homepage screens use synthetic data in the fictional Alder Cove Energy LLC demo tenant. ## AVEVA PI connectors Sightwell connects to systems you already run. Historians, asset models, SharePoint, SQL and spreadsheets stay where they are. Named on the site: - AVEVA PI - Asset Framework - AVEVA Historian - Ignition - OPC UA - SharePoint - File shares - SQL - Excel - Power BI The architecture also names CMMS and ERP among the systems it reads. Connectors are read-only. See [Read-only gateway boundary](#read-only-gateway-boundary). ## Logic Discovery Sightwell finds calculations that have become operational facts: spreadsheets, PI calculations and dashboards with hard-coded assumptions. It traces a figure back to the plant data underneath, including where the same constant has been copied. It surfaces constants, thresholds and duplicated formulas. It does not approve a change. Your engineers decide what changes. The homepage calls this discovery and governance, not automatic approval. ## P&ID integration Drawings sit in SharePoint. Tags sit in PI. Sightwell links each instrument on a P&ID to its PI Asset Framework element, so an investigation can follow the process as well as the numbers. A drawing and the historian can disagree; the product is built to show that gap rather than hide it. ## Report lineage A number on a dashboard or in a board pack can be traced through the workbook, the query and the dataset back to the historian, and compared with it. You keep Power BI, Excel and the other reporting tools you already use. ## Pre-submission checks Before a calculated report leaves the organisation, Sightwell compares the method in the workbook with the method an approved document requires, and can rebuild the figure from historian tags. The homepage uses an emissions return as the worked example. The same check applies to the class of document named there: emissions returns, customer quotes and contract statements. Sightwell shows the gap. People correct the report. ## From signal to service A maintenance question can be followed from the signal, to the drawing, to manufacturer guidance, to the maintenance schedule, to a scoped partner, and to a decision. The partner works from the case. They do not receive a connection to the historian. ## Evidence Each claim in an answer links to the tag, time range, document page or formula it came from. When the sources do not support a conclusion, Sightwell says so and shows what is missing. The standard on the homepage: no evidence, no confident answer. ## Partner access An equipment vendor can investigate machines that are in its contract without a connection to the plant network. The same question, asked by a different identity, is answered from a different permitted context. Authorization is applied before retrieval, so out-of-scope assets are not sent to the model. ## Bring Your Own AI. Keep it out of the control loop. ChatGPT, Claude, Sightwell AI or your internal agents consume the same governed industrial context. Sightwell sits between those clients and your infrastructure. MCP (Model Context Protocol) is the open standard Sightwell implements, so a compatible client can connect to a customer's deployment. The value underneath that interface is the connectors, asset relationships, normalized context, permissions, evidence, documents, time series and audit. The public website is not that MCP endpoint. A customer's MCP endpoint is part of their own Sightwell deployment. This origin publishes an empty catalog at `/.well-known/ai-catalog.json` and `/.well-known/mcp/catalog.json` so a client looking for a server on sightwell.ai does not invent one. ## Read-only gateway boundary Sightwell starts as a read-only intelligence layer. There is no write path from Sightwell to PLC, DCS, SIS or operational control. Those systems keep running independently of Sightwell. AI is not in the control loop. ## Azure tenant deployment Sightwell runs where the industrial data already belongs. There is no mandatory SaaS environment, and plant data is not uploaded to Sightwell. | Deployment | Status | Where it runs | | --- | --- | --- | | Customer Azure | Available | Inside your Azure tenant, under your subscription and policies, with Microsoft Entra ID and private connectivity to the plant | | On-premises | Available | Kubernetes in infrastructure you control, virtualized or bare metal, next to the systems it reads | | Hybrid | Available | Collectors and processing placed between plant, edge and cloud where your policy allows | | AWS | Planned | Not available yet | | Google Cloud | Planned | Not available yet | Also named on the site: private networking, encryption, RBAC / ABAC, audit trails and data sovereignty. Getting started is deploy, connect (historian, SharePoint and identity, read-only), discover (assets, documents and calculations), then ask. Timelines depend on site scope and data readiness. ## What this website is sightwell.ai is the marketing site. It does not host plant data, and it does not accept MCP connections. Product examples are replayed synthetic data for the fictional Alder Cove Energy LLC demo tenant. Sightwell Connectivity, at [/connectivity](https://sightwell.ai/connectivity), is the iPhone, iPad and Mac app. It connects to an organisation's own Sightwell server. The [website privacy notice](https://sightwell.ai/privacy) covers sightwell.ai only. Trademarks named on the site, including AVEVA, PI System, Ignition, Microsoft Azure, Entra ID, SharePoint, Excel, Power BI, AWS, Google Cloud, ChatGPT and Claude, belong to their respective owners. Sightwell is not affiliated with or endorsed by them.