> ## Documentation Index
> Fetch the complete documentation index at: https://docs.speckle.systems/llms.txt
> Use this file to discover all available pages before exploring further.
> ## Agent Instructions
> These docs describe Speckle as it works today, including everything released in 2026.9. Treat page content as current unless the page or section says it applies only to versions published before 2026.9.
> Content scoped to versions published before 2026.9 (the object graph and root object, Proxies, Receive2, operations.receive, object.children) applies only to model versions in that earlier data format. For versions published in 2026.9, answer from the Object model pages and the SDK migration guides.
> The What's new pages summarise what changed in 2026.9 and link to the page that documents each change. Answer from the linked page rather than from the summary. Legacy (v2) connectors are retired; send questions about them to the community help forum.
> Match guidance to the customer's deployment, model data format, connector, plan and permissions. Self-hosted servers on a release before 2026.9 and compatibility-mode workflows follow the Compatibility mode page. If an unknown version or data format changes the answer, ask one focused question; do not infer the deployment version from the date alone.
# Assess model changes with Speckle Intelligence
> A worked example — ask Chat what changed between two versions, read the summary, then filter and colour the model by how significant each change is.
Speckle Intelligence can detect what changed between two versions, explain it in plain language,
and — because Chat can also drive the viewer — colour the model by how significant each change is.
There is no separate diff view: you ask. This page walks through one real conversation end to end,
from the first question to a scene coloured by change severity, so you know what to expect and can
reuse the pattern on your own models.
On a self-hosted Speckle Enterprise Server, your server administrator turns on Intelligence and
connects it to your organisation's AI provider. See [Enable
Intelligence](/developers/server/deployment/enterprise-license#enable-intelligence).
## What this workflow does
* **Detect** — find the two versions to compare and diff them, without you opening a versions
panel or clicking a diff button.
* **Report** — explain what changed in plain language: scale first, then the categories that
carried the change.
* **Assess impact** — score changed elements by how significant the change is, not just
changed-or-not.
* **Visualise** — colour the model by that score so the impact is visible, not just written down.
## Walkthrough
In Chat or the viewer sparkle, ask a plain question — no version pickers, no diff button:
> What is the most significant change between the last two versions of the model?
Intelligence lists the model's version history, attaches the previous version alongside the
one you have open, and runs a diff between them. You do not see those as separate steps; they
happen as part of answering the question.
The reply is a written summary, not a raw diff dump — scale first, then the categories that
carried the change, in order of significance. On one structural model, the answer read:
> The most significant change is a structural steel re-specification that swept through the
> model. Scale: 6,598 elements modified, 119 added, 105 removed across 66 levels — only \~60%
> of the model was left untouched. Structural Framing (3,873 modifications) shows beam
> families replaced en masse, with cross-sections cleared and rebuilt and the material grade
> upgraded (Steel 43-275 → Steel 345 MPa). Structural Columns (135) and Structural Connections
> (376 modified, 61 added, 22 removed) shifted to match. Walls (1,753 modified) look like a
> secondary effect of the structural resizing rather than an independent edit.
Intelligence reads this from the live model — the same [connected, queryable
data](/analytics/intelligence#how-it-works) Chat uses for any other question — not from a
static export or a change log someone maintained by hand.
Keep going the way you would with a colleague:
> Filter to just those elements and colour them by how significant.
You do not have to define "significant" or set a threshold. Intelligence scores each changed
element by how many of its properties moved, buckets that into tiers, and applies a filter and
colour-by in the viewer you are already looking at — the same [Filters
panel](/3d-viewer/exploration#filter-and-color) documented for manual use.
The reply names the tiers it chose and what landed in each one:
> 🔴 Major (9+ params changed) — 1,757 elements, the deepest re-specs. 🟠 Moderate (5–8 params)
> — 2,587 elements. 🟡 Minor (1–4 params) — 2,254 elements. 🟢 Added — 119 new elements.
The scene is now filtered to only the changed elements, coloured by tier, so the heaviest
re-specification is visually obvious without you building that filter by hand. From here you
can ask to isolate one tier, break it down by level, or start a new question — the filter
stays until you clear it or ask for something that replaces it.
## Turn this into a skill
If your team asks this shape of question often — after every structural issue, before every
coordination meeting — save the pattern instead of retyping it. [Save the chat as a
skill](/analytics/intelligence-skills#save-a-chat-as-a-skill) once the conversation behaves the
way you want, or write the significance tiers and vocabulary into a skill by hand so every
comparison uses the same thresholds and the same added/omitted/revised language. See [Reuse
organisation playbooks with Intelligence skills](/analytics/intelligence-skills).
You can also save the conversation as a [Report](/analytics/reports) if you want a shareable
artefact rather than a one-off chat — bind the two versions as source aliases and the comparison
becomes something you or a teammate can reopen without re-asking Intelligence.
## FAQ
No. Ask about "the last two versions", or name one if you want an older baseline — for example
"compare this to the version from before the coordination meeting." Intelligence finds and
attaches the versions it needs.
Not fixed. In this example Intelligence scored elements by how many properties changed and split
them into rough tiers. Ask it to explain its thresholds, or tell it to weight geometry changes
over parameter-only changes, if you want a different definition of significant for your project.
Yes. Like every other Intelligence query, comparing versions needs both versions in the data
format introduced in 2026.9. Versions published before 2026.9 can't be queried; publish again
with a 2026.9 connector. See [Speckle's new data model](/workspaces/data-model-migration).
Yes — the questions on this page are plain language, not a script. Ask them in Chat or the
viewer sparkle on any project with at least two versions in the 2026.9 data format.
The dedicated version-diff view was removed from the 3D viewer. Asking Intelligence, as on this
page, replaces it. See [Compare versions](/3d-viewer/compare-versions).
## See also
* [Ask questions about your models with Speckle Intelligence](/analytics/intelligence)
* [Create and share Intelligence Reports](/analytics/reports)
* [Reuse organisation playbooks with Intelligence skills](/analytics/intelligence-skills)
* [Compare versions](/3d-viewer/compare-versions)
* [3D viewer](/3d-viewer/introduction)
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.