AI assistant for Smallworld Geo Network Management

Making AI Real!

Every answer shows its source.

SmartGuide works inside Smallworld Electric Office. It answers from your own manuals and shows you the page it took the answer from — then turns plain questions into real queries, traces and edits against your live database.

Book a walkthrough See what it does

Runs inside your GNM session Your infrastructure or ours Not tied to one AI provider

Smallworld Electric Office — SmartGuide Session

Welcome to SmartGuide. Your selected profile is EO. Your session has started — how can I help?

Source · Electric Office documentation

Designing the line extension

To complete this design, you can follow these steps:

Create a design project
Create a Service Point at the building location
Create poles to support the new MV line
Create a tap to connect the existing line to the line extension
Create the single phase MV line
Add a single phase transformer
Create a LV line from the transformer to the house
Create a fuse on the downstream side of the transformer

Ask SmartGuide

The reality today

Three things slow every GNM team down

None of them are new. All of them are expensive, and all three are answerable from knowledge your organisation already owns.

Knowledge

The documentation exists. Nobody can find it.

Manuals, work instructions, project provisions and third-party documents live in different places. When someone is stuck mid-task, searching four systems is not an option — so they interrupt a colleague instead.

Access

The data exists. Only specialists can reach it.

Answering "which poles in this area predate 2020?" means knowing the schema and writing a query. The people who most need the answer are usually the people who cannot write one.

Fit

Generic AI does not know your network.

An assistant that has never seen your asset types, your topology or your data model produces answers that are plausible and wrong. In a network environment, plausible and wrong is the dangerous combination.

Answers you can check

It searches from all of your documentation and displays the answers

SmartGuide reads the manuals, data model, work instructions and process documents you already have. Every answer carries a source marker. Click it and the actual documentation page opens, with the passage the answer came from.

>How do I make a house connection?

→A fourteen-step procedure in plain language, in seconds — with a source marker on the steps that came from your documentation.

>What does the validate schematic check actually test?

→An explanation drawn from your own material, not from a model's general knowledge of GIS.

Grounded in your material. Technical documentation, application manuals, work instructions, project-specific provisions and third-party documentation.
Every answer is checkable. SmartGuide answers from what you have documented and shows the source, so nobody has to take the answer on trust.
New starters stop queuing. Guidance in plain language without needing to know who to ask or where to look.
Nobody leaves the application. The answer arrives in the panel beside the map, not in a browser tab.
Ask out loud. Speech-to-text for questions and text-to-speech for answers, for hands-busy work.

Ask the live database

Questions become real queries, traces and edits

SmartGuide translates what someone wants to know into the right Smallworld query, runs it against the live session and puts the results in the Explorer dialog. No Magik. No schema knowledge. Every prompt below is one we run in the product.

>Find all poles within my selected area made before 2020 and trail them

→Builds the query, returns the matching poles in the Explorer dialog and sets them as the active trail on the map.

>Find all poles within 500m of my current trail point and select them

→Runs a proximity search from where you are standing on the map and marks the results as the active selection.

>Do a 500m trace out

→Network traces from the current selection — trace out, shortest path and proximity, plus an advanced electrical trace with direction, phasing and circuit parameters in Electric Office.

>Open the Quality Manager

→Runs your quality tests the same way the Quality Manager does, then summarises the flags in plain language instead of leaving you to read them.

>Change 'date removed' to today on every pole in my selection

→Shows an old-value / new-value table for every affected object and waits for you to approve it. Nothing is written until you say yes.

>Write some Magik that prints all my selected objects

→Generates the Magik, explains what it does and gives you a button to run it in the session. You read it before it runs.

ⓘ The decision stays with the person. SmartGuide proposes the query, shows the results and asks before it writes. It increases throughput by removing the mechanics, not by removing the operator.

Beyond Smallworld

One question, answered across your systems

The full picture is rarely in one place. SmartGuide reaches document stores, asset management and enterprise systems through MCP, so a single question can draw on all of them without anyone switching tools.

Connections are configured per deployment against the systems you actually run — SharePoint, SAP, EAM, document repositories and SQL sources among them.

>Query the external server for all poles marked as joint use, then select those IDs in my session

→SmartGuide fetches the joint-use asset IDs from the external system, matches them against the live Smallworld database and highlights them on your map — one question, two systems, no export.

ⓘ No vendor lock on the model. The language model behind SmartGuide can be changed to suit your organisation's preference or data-residency requirements.

Data quality

It tells you when the data is the problem

Most assistants assume your records are correct. Asset data quality issues quietly compromise connectivity models, calculations and operational decisions — so SmartGuide checks, and flags what it finds.

Flagging

Topology and connectivity gaps

Topology errors, missing connections and connectivity gaps are flagged so somebody can act on them, rather than discovering them downstream in a calculation.

Validation

Chain validation

Checks whether objects are genuinely connected, whether field values are plausible, and what the object history shows.

Conflict

Conflicts found before the merge

When two people have edited the same objects, SmartGuide identifies the conflict, retrieves the affected objects and shows who changed what and when — before a merge, not after.

Security

What the model sees, and what it never touches

Queries run against your live database and the results stay in your session. The language model receives filtered context — never your GIS database, never raw files.

Live data stays put. Network data is never extracted or sent to an external system. Queries run against your database and results stay in your session.
Profiles set the boundaries. User profiles control what each person can see, query or act on. A field technician and a data manager work within different limits.
Nothing is written unasked. Bulk edits are previewed as an old-value / new-value table and applied only on approval.
Every conversation can be logged. Monitoring gives your organisation visibility of how AI is being used across teams.

Hosting

You decide where it runs

SmartGuide supports three options for deployment.

On your infrastructure

Hosted by you

Runs entirely in your environment
No data leaves your network
Full control over security and access

Managed

Hosted by Realworld

Hosting and setup handled by us
Support and maintenance included
Shared instance

Cloud

Hosted in the cloud

Managed on AWS or a comparable service
No on-premise hardware

Diagnostics

See what the AI is actually doing

Rolling out an assistant is a governance question as much as a technical one. Where Realworld Diagnostics is in place alongside SmartGuide, it turns the assistant's activity into something a manager can read.

A trail of edits and versions. User edits and actions are queryable, so managers can see what changed and who changed it.
Knowledge gaps, found for you. Questions that return weak answers point straight at the documentation you are missing.
Usage, cost and impact. A clear picture of how the assistant is used across the organisation and what it is worth.
Quality ratings and safety checks. Answer quality is monitored rather than assumed.

How it works

Retrieval first, generation second

SmartGuide is a retrieval-augmented system. It finds the relevant material in your knowledge base first, and only then asks a language model to put it into words — which is why an answer can always point back at a source.

01The questionAsked in plain language, typed or spoken, from inside GNM.

02Transformed queryTurned into a retrieval query, and where relevant a Smallworld query.

03RetrievalThe relevant documents and records are found in your knowledge base.

04AnswerThe model writes a clear answer from the retrieved material.

05SourcesThe answer is returned with the references it was built from.

06MonitoringQuality, safety and usage are tracked through Diagnostics.

Built on an enhanced RAG stack with a custom proxy layer, an optimised retrieval pipeline and monitoring — embedded in the Smallworld interface as an HTML5 component, so it looks and behaves like part of the application.

Choosing an assistant

What to ask any AI assistant for GNM

Most assistants in this space can run a trace and edit in bulk. These are the questions that separate them.

The question to ask	SmartGuide	A typical GNM assistant
Can it answer from our own manuals and show me the page?	Yes — a source marker on the answer opens the documentation	Usually context-aware help only
Is knowledge and action one product or several?	One assistant, one deployment	Often split across separate products
Can it run entirely on our own infrastructure?	Yes — client-hosted, Realworld-hosted or cloud	Commonly vendor-hosted with encryption promises
Does it tell us when our data is wrong?	Flags topology, connectivity and conflicts before a merge	Generally assumes the data is correct
Can we change the underlying AI model?	Yes, to suit preference or data residency	Increasingly yes — worth confirming
Does it show us which documentation is missing?	Knowledge-gap detection from weak answers	Rarely
Can people ask out loud?	Speech to text and text to speech	Typically typed only

The right-hand column describes patterns common across assistants in this category, not any one named product. We would encourage you to put the same questions to us and to anyone else you are evaluating.

Questions

The things people ask first

Can SmartGuide make something up?

SmartGuide answers from the material in your knowledge base rather than from a model's general knowledge, and it shows you the source it used. That is what makes an answer checkable: if something looks wrong, you can open the source and see for yourself. Where the knowledge base does not cover a question, and Diagnostics is in place, it records the gap so you know what to document next.

Does our network data leave our environment?

No. Queries run against your live database and the results stay inside your session. The language model receives filtered context — the question, matching document passages, the schema fields it needs and the rows your query returned. It does not receive your GIS database, whole documents or raw files. If you host SmartGuide yourself, nothing leaves your infrastructure at all.

Do our users have to learn Magik or the schema?

No. That is the point. Someone asks for what they want in ordinary language and SmartGuide works out the query, runs it and puts the results in the Explorer dialog. It knows your asset types, your topology and your data model.

Can it change data by accident?

Edits are proposed, not performed. A bulk change shows an old-value / new-value table covering every affected object, and nothing is written until a person approves it. Generated Magik is shown to you with an explanation and runs only when you choose to run it. User profiles limit what each person can act on in the first place.

Which AI model does it use?

Whichever suits you. SmartGuide is not tied to a single provider, and the underlying model can be changed to match your organisation's preference or your data-residency requirements.

What do we need to have in place before starting?

Two things: the documentation you want SmartGuide to answer from, and a decision about where it runs. Setting up is building the knowledge base from your material and configuring the chat system. Your documentation does not need to be tidy first.

How does SmartGuide work with other Realworld products?

SmartGuide stands on its own — no other Realworld product is required to run it. Where Diagnostics is also in place, the two work together: SmartGuide can reference what Diagnostics has recorded, so an answer can take account of activity carried out by you and by others across your company. Diagnostics enhances SmartGuide rather than being a prerequisite for it.

How does it fit with our existing Smallworld setup?

SmartGuide is embedded in the Smallworld interface as an HTML5 component, so it appears as a panel inside the application your teams already use. It can also be opened as a pop-up window that stays connected to the same session, for people who prefer it beside the map rather than docked in it. And it can reach systems outside Smallworld — document stores, asset management and enterprise systems — so a single question can draw on all of them.

Available 31 August 2026

See it answer a question from your own documentation

Bring a question your team actually asks and a manual you would like it answered from. That is a more useful test than any demo script we could write.

Book a walkthrough