Connect AVEVA PI to Copilot
You're building a Copilot agent that should be able to answer questions about plant operations — and the plant's process data lives in AVEVA PI. The agent needs to retrieve tag values, pull historical trends, and surface process context on demand. The problem is that AVEVA PI was designed for SCADA systems and historians, not AI agents — and getting a Copilot agent to call PI data reliably requires bridging an OT network gap that Copilot Studio has no built-in answer for.
If you've already tried something and hit a wall, the list below is probably why. The barriers aren't obvious until you're in the middle of them.
Why it's hard
Seven reasons connecting a Copilot agent to AVEVA PI is harder than pointing a Custom Connector at it.
Copilot agents call HTTP endpoints — not OT protocols
Copilot Studio agents access external data through Custom Connectors, which make outbound HTTP calls to REST APIs. They have no concept of OPC, PI SDK, or any OT protocol. The only interface PI exposes that an agent could theoretically call is PI Web API — but that's a separate component that needs to be deployed, configured, and authorized before it exists at all.
PI Web API is on the OT network — Copilot Studio calls from Microsoft's cloud
When your Copilot agent invokes an action, Microsoft's Power Platform infrastructure makes the outbound HTTP request. That request originates from Microsoft's cloud — not your corporate network. A PI Web API instance sitting on a private OT VLAN is completely unreachable from there. Getting a network path between Microsoft's cloud and your OT historian requires deliberate, security-reviewed network engineering.
PI tag names are meaningless to an AI agent without context
PI tag names follow internal naming conventions — strings like "CDT158.PV" or "UNIT2.FIC203.OUT" that mean nothing outside the engineering team that configured the historian. A Copilot agent that queries raw PI tags will either return cryptic identifiers the user can't interpret, or require an expensive system prompt full of tag documentation that grows every time the historian is reconfigured.
Raw PI time-series data overflows action response limits
PI can record at sub-second intervals across hundreds of tags. Even a modest time-series query — a handful of tags over a few hours — returns thousands of data points. Copilot Studio action responses have payload size constraints, and returning bulk time-series data will exceed them. The agent needs tools that return aggregated, summarized responses — which requires aggregation logic somewhere in the chain.
Power Platform Custom Connectors require IT admin approval
In most enterprise Microsoft 365 tenants, custom connectors need to be reviewed and approved by a Power Platform admin before they can be used in production agents. If the connector points at an OT-adjacent endpoint, the security review process can take weeks. The connector also needs to clear change management, especially if it's the first external data connection the team has made from Power Platform.
Copilot for Microsoft 365 and Copilot Studio are different products
Many teams conflate these. Copilot for Microsoft 365 — the AI assistant in Teams, Outlook, and Word — doesn't support custom data connectors for plant data. Building an agent that can query AVEVA PI requires Copilot Studio, which is a separate licensed product with its own environment, its own admin requirements, and its own per-message or per-user pricing model.
PI AF context isn't available in the cloud — and agents need it
PI Asset Framework (AF) is what gives raw PI tags their business context — descriptive names, engineering units, asset hierarchy, and normal operating ranges. Without AF context, an agent returning a PI value has no way to tell the user what it means. AF is an on-premise component; surfacing its context alongside raw values in a cloud AI agent requires additional work beyond just connecting to PI Web API.
What teams usually try
Three common approaches — and where each one breaks down.
Custom Connector pointed at PI Web API directly
Deploy PI Web API, expose it through a reverse proxy or DMZ, define a Copilot Studio Custom Connector using its OpenAPI spec, and wire the agent's actions to call it. The most direct path on paper.
The catch: Exposing PI Web API to a cloud-accessible endpoint requires significant network and security engineering. The PI Web API OpenAPI schema is complex and not designed for AI agent consumption. Raw PI responses need transformation before an agent can use them. And every PI server upgrade risks breaking the connector schema.
Custom middleware API that bridges PI to the cloud
Build a custom REST service that sits between PI and the internet — it queries PI internally, transforms the responses, and exposes a clean API that Copilot Studio's Custom Connector can call. Gives full control over the schema.
The catch: You've become the middleware team. The custom service needs to be built, hosted, secured, monitored, and maintained. Every PI schema change or network update that touches the middleware is another engineering task. And you've only solved AVEVA PI — the next historian on the plant is a separate build.
Periodic export to SharePoint, agent reads files
Export PI data to CSV or Excel files on a schedule, upload them to SharePoint, and configure the Copilot agent to search and read those files as its knowledge source. Avoids the network connectivity problem entirely.
The catch: The agent is working from stale snapshots, not live data. "What is the current temperature on Tank 3?" becomes "What was the temperature on Tank 3 as of the last export?" The agent can't answer real-time questions, and the export pipeline needs to run reliably to keep the files current.
How TrendOps solves it
TrendOps Edge connects to AVEVA PI on the OT network and forwards normalised, context-rich data to TrendOps Platform in the cloud. Your Copilot Studio agent calls TrendOps Web API through a single Custom Connector — no OT network access required, no PI SDK, no custom middleware to build.
Your Copilot agent calls TrendOps Web API through a single Custom Connector. The endpoint is cloud-accessible. No reverse proxy, no DMZ, no firewall exception pointing at your OT network.
TrendOps resolves PI tag names and AF context before data reaches the agent. The agent works with descriptive tag names and receives values with units — not raw numeric strings it can't interpret.
TrendOps handles aggregation server-side. The agent receives summarised data appropriate for a conversation — not thousands of raw PI compressed values that overflow action response limits.
TrendOps Edge handles all PI connectivity on the OT side. The Custom Connector calls a standard REST API — no AVEVA licensing, no PI client software, no on-premise SDK dependencies on the cloud side of the integration.
TrendOps Edge polls AVEVA PI continuously and keeps TrendOps Platform current. TrendOps keeps process data current in the cloud, so the agent gets live values — not a snapshot from last night's export job.
TrendOps Edge pushes data out over a single outbound MQTT connection. The OT network has no inbound path to the cloud — the Copilot agent never reaches into the plant network.
Because TrendOps normalises data from all connected sources, the same Custom Connector and the same tools work whether the underlying data comes from AVEVA PI, DeltaV, FactoryTalk, or SQL. The agent doesn't need to know which historian a tag comes from.
The Copilot Studio agent can be published to Microsoft Teams, SharePoint, or as a standalone web chat — wherever your operations and engineering teams work. No separate app to build or maintain.
What you end up with
A Copilot agent that can answer live questions about AVEVA PI data — deployed in Teams or SharePoint, with no OT network exposure in the call path.
The agent calls a standard REST API — no AVEVA client software on any cloud component.
Current process values, not file exports — the agent knows what's happening right now.
Copilot calls TrendOps cloud API — the OT network is never in the agent's call path.
One connector, every OT historian — the agent isn't limited to AVEVA PI.
Don't want to build your own AI agent integration?
TrendGuru — TrendOps's built-in AI agent — already has access to your plant data. Ask questions about your AVEVA PI historian in plain language and get answers instantly, deployed directly in Microsoft Teams. No custom connector to configure, no OpenAPI spec to write, no Power Platform admin approval to wait for.
Learn about TrendGuru →Your Copilot agent shouldn't need a PI developer on standby to answer a question about plant data
TrendOps gives your Copilot agent a clean, cloud-accessible API for AVEVA PI data — so it can answer live process questions without touching the OT network.
Book a Demo