Integrations and data sources
The two kinds of connection Edmund makes to systems you already have: an integration that synchronizes documents from SharePoint on a schedule, and a data source that queries an operational database live at the moment you ask.
Not everything Edmund knows has to be uploaded into it. It also connects to systems you already have, in two ways, and the difference is what each does with your data:
An integration synchronizes on a schedule and its content is indexed in Edmund. SharePoint is the one that exists today: leave the documentation in its libraries and each project picks the folders Edmund reads for that machine.
A data source is queried live. Edmund writes SQL against the schema and fetches the answer at the moment you ask — stock, spare parts, fault history — and nothing is copied in.
Both are managed for the whole organization under Connections in the left navigation, which opens into Integrations and Data sources, and assigned to the projects that should use them.
What you already have is listed here; when there is nothing yet, Add connection is how the first one is made.
Adding either is an administrator task. Everyone can see what a project uses, but only an Administrator can add or change a connection.
Connect SharePoint
Under Connections, choose Add connection. The next page lists what can be connected, grouped by how it behaves.
The two groups, and the sentence under each saying what it does with your data. Pick SharePoint. Edmund asks for the directory (tenant) ID, the application (client) ID and the client secret from the app registration in Microsoft Entra ID, and the document library it reads from. Then Connect and continue.
Edmund checks the details against the library before saving — a failed check saves nothing.
Edmund asks for read-only application access. It reads the library, never writes to it, and never acts on behalf of a person. Give it a registration scoped to no more than that.
Which folders a given machine uses is decided per project, so a technician asking from one project gets that machine’s documentation rather than everything the plant has on SharePoint.
Connect a database
In Organization settings, open Connections. The overview lists every connected database and whether it is reachable.
Click Add connection.
Pick the database type. Which types appear depends on your organization’s feature flags — PostgreSQL, Microsoft SQL Server, Oracle Database, InfluxDB or Snowflake.
Fill in the connection details — Connection name, Host URL, Database name, and the Username and Password of a read-only account — then choose Next.
In the wizard’s second step, Project assignment, pick the project (or projects) that should query the source, and save. Each project’s settings page shows its assigned sources under Data sources.
Add a description
Each database can carry a Database description — a field on the connection form, also reachable later by clicking the database in the data-source overview. It’s optional, but a well-written description markedly improves query accuracy — describe what the tables hold and how they relate, so Edmund knows where to look.
A description can sit at either of two levels, and the difference is worth knowing:
On the connection — global. It applies wherever that database is queried, from any project.
On the project — local to one machine. Open the project’s settings and use Manage description. It applies only there.
The local one is how you keep an answer to the machine in front of you. Describe the data as it relates to this line, and a technician asking from that project gets what concerns their machine rather than the whole plant.
Spend the time on the description. It’s the single biggest lever on how well Edmund reads your database.
Connect with a read-only account
Connect with a database account that only needs to read the data Edmund should see. Read-only access is the safe practice — it keeps queries from ever changing your data.
How database answers look
A database answer comes back as a results table plus a short summary, pulled live at query time — not as Source links, which belong to document and journal answers. See Reading answers & citations.
And when a table isn’t enough, ask for a chart — Edmund draws it from the query result right in the chat, with Export PNG to take it away: