AI for CAD Tools
Onshape ships AI Advisor, FeatureScript generation and AI search. Here is what each one reaches, and where the API is still the only door into your documents.
·
⏱
8 min read
Michelle Ben-David
Michelle Ben-David is a mechanical engineer and Technion graduate. She served in an IDF elite technology and intelligence unit, where she developed multidisciplinary systems integrating mechanics, electronics, and advanced algorithms. Her engineering background spans robotics, medical devices, and automotive systems.
BOTTOM LINE
Onshape’s built-in AI is real and it is narrower than the label suggests. AI Advisor is a guidance layer grounded in Onshape’s documentation, and it shortens the path from an unfamiliar problem to the accepted way of solving it. FeatureScript generation reaches further, into authoring custom features, with the vendor’s own reminder that generated output still needs engineering review. Everything structural about your actual designs, document contents, workspaces against versions, assembly relationships, release state, and permissions, still comes through the API, which on an API-first platform is a clean answer rather than a workaround. And the material outside Onshape, in PDM, PLM, ERP, standards, and network drives, needs a layer above the CAD platform entirely.
Onshape has shipped a longer run of AI features than most engineers realize. There is a guidance assistant that sits in the design environment, language-driven FeatureScript authoring, model search that reads generated descriptions instead of filenames, text-prompt rendering, and machine-learning transfer of drawing annotations. PTC opened Onshape Labs in July 2026 as an early-access home for the newer work, and a FeatureScript MCP Server reached general availability there in August 2026.
None of that answers the question an engineer actually asks when a vendor says AI: will it tell me something about my design that I did not already know? Onshape is an interesting case here because it is API-first by construction. Anything outside the browser reads a document through the REST API, which makes the line between what the built-in AI handles and what needs a programmatic path unusually clean. This post walks that line feature by feature, then looks at the part neither side covers.
What Onshape actually ships as built-in AI
Five capabilities are live today, and they reach in very different directions:
AI Advisor, released on October 15, 2025 and built on Amazon Bedrock, gives real-time guidance in the design environment: step-by-step recommendations, troubleshooting help, and best practices drawn from the Learning Center and the help documentation.
LLM-powered FeatureScript generates and autocompletes FeatureScript code, so a custom feature no longer requires someone on the team who writes it fluently.
AI-powered search for models matches against AI-generated descriptions of models rather than filenames and metadata alone.
AI Quick Render applies image generation models to produce presentation-grade renderings from a text prompt.
Replicate Annotations uses machine learning and probabilistic matching to transfer annotations between drawings, including dimensions, tolerances, layout, and placement.
Onshape also describes AI agents as work in progress, with a stated ambition covering design feedback, questions about model data, deliverable export, failing-feature repair, parameter updates, and goal-driven geometry changes. Those are roadmap items, not shipped behavior, and it is worth keeping the two categories apart when a plan is being built around them. The same split shows up across the rest of this cluster: the comparison of built-in AI across CAD and PLM platforms puts the vendors side by side, and the deep dives on what Creo ships and what Siemens NX ships follow the same pattern of a live documentation assistant plus a set of narrower, more experimental tools behind it.
IN PRACTICE
It integrates directly with PLM and existing workflows, making past designs, standards, and calculations instantly available. The result is fewer errors, faster decision-making, and a more consistent process across teams.
- Sergey G., Board Member
AI Advisor answers questions about Onshape, not about your model
This is the single most useful distinction to hold onto. AI Advisor is grounded in Onshape’s own material: product documentation, the Learning Center, and accumulated best practice. That makes it genuinely good at a specific class of question. Why does this loft keep failing, what is the accepted way to build a configurable part family, which mate type belongs here, how should a drawing template be set up. An engineer three weeks into the platform gets real value from it, and so does an experienced one working in a corner of the product they rarely touch.
What it is not is a layer that reasons over your design data. It does not know that the bracket in front of you was revised twice last quarter because a supplier changed the fastener, or which of your assemblies already use the bearing you are about to specify, or what the release state of the sub-assembly two levels down happens to be. Those answers live in document structure and project history, and a documentation-grounded assistant has no path to them. The distinction matters because vendor messaging tends to collapse it: guidance about the tool and intelligence about the work get presented under one label, and only the first one has actually shipped.
Read the boundary that way and AI Advisor stops being disappointing and starts being useful for what it is. It shortens the time between hitting an unfamiliar problem in Onshape and knowing the accepted way to solve it. That is a real saving, repeated many times a week. It is simply not the same product as an assistant that answers questions about your parts.
FeatureScript generation is the deepest reach the built-in AI has
FeatureScript is Onshape’s language for building custom parametric features, and it is where the built-in AI gets closest to real geometry. The in-product generation and autocomplete lower the cost of writing a custom feature. The FeatureScript MCP Server, announced when Onshape Labs launched in July 2026 and generally available inside Labs from August 2026, goes further: it connects external language models to FeatureScript over the Model Context Protocol, giving a text-to-code-to-CAD path where a natural-language instruction becomes a reusable custom feature that can be shared across teams and projects.
That is a meaningful capability, and it is worth being precise about its shape. It is authoring, not reasoning about designs that already exist. It produces a feature definition that then runs deterministically, which is the right architecture for CAD automation, because the reviewable artifact is code rather than a one-off geometric result. It is also scoped to FeatureScript specifically, not to general document read and write, and it sits in an early-access environment rather than in the mainline product.
PTC’s own framing is the useful caution here: AI-generated outputs and automation still require engineering review, to confirm the resulting designs meet functional, manufacturing, safety, and regulatory requirements. A generated custom feature is a draft of an engineering tool, and it inherits every obligation that comes with one. Teams that treat it as a starting point for review get the productivity gain. Teams that treat generated automation as validated by the fact that it ran will find the problem later, in a place where it costs more.
Where the API still wins
Because Onshape is API-first, the things the built-in AI cannot answer are not mysteries. They are documented, queryable, and sitting one REST call away. The structural facts an engineering question usually depends on are exactly the ones the API hands over as data:
What a document reference actually points to, given that a document is a container holding Part Studios, Assemblies, drawings, and other element types rather than a single file.
The difference between a workspace, which is an editable branch, and a version, which is an immutable snapshot, plus the automatic checkpoints in between.
Assembly structure, including part relationships and mate connections, retrieved as queryable data without opening the file.
Release state, which is a workflow fact and entirely separate from whether a document exists.
Permission boundaries, including the way a linked document can grant partial access independently of direct permissions.
We covered that surface in detail in what the Onshape API hands an AI agent, so it is worth reading alongside this one rather than repeated here. The point for the built-in-versus-API question is narrower. Every one of those five is a precondition for a trustworthy answer about a real design, and none of them is something AI Advisor will tell you. Add cross-document work, such as finding every assembly in a few hundred documents that consumes a given part, and the gap widens further: that is an indexing and retrieval job, and it needs programmatic access to build the index in the first place.
The gap neither one closes: everything outside Onshape
Here is where the honest accounting gets uncomfortable for both halves of the argument. AI Advisor is grounded in Onshape’s documentation. The API is grounded in Onshape’s documents. Neither one reaches the material that most engineering questions actually depend on, because that material was never in the CAD platform to begin with.
Think about what a real question needs. Released revisions and approval history sit in a PDM or PLM system. Supplier terms, lead times, and cost sit in the ERP. The applicable clause sits in a standard. The load case someone ran three years ago sits in a spreadsheet on a network drive, and the reason the design went the way it did sits in a review document nobody has opened since. An assistant confined to the CAD platform, however well it reads that platform, answers a fraction of the question and leaves the engineer to reassemble the rest by hand.
Closing that gap is what an AI intelligence layer above the design tools is for, and it is where Leo sits. Leo connects to an organization’s full knowledge base: PDM, PLM, local and network directories, and ERP, with integrations for leading PDM and PLM platforms including SolidWorks PDM, Autodesk Vault, PTC Windchill, Siemens Teamcenter, Arena PLM, and others. It already ingests Onshape document links, so the CAD platform becomes one source among several rather than the boundary of what can be asked. Because it runs on a Large Mechanical Model trained on more than a million pages of standards, books, and technical articles, the answers come back in engineering terms and carry a citation that can be opened and checked against the source. It is an intelligence layer on top of those systems, not a replacement for any of them, and it is SOC-2 certified and GDPR compliant, with no AI trained on customer data. For the narrower question of which tools work alongside the platform day to day, we keep a current view in our roundup of the best AI tools for Onshape.
FAQ
Answers Onshape’s AI Cannot Reach
Connect your PDM, standards and past designs to one AI layer.
Leo sits above Onshape and your PDM to surface released revisions, standards and past decisions your CAD session cannot see, each answer with a citation you can verify.
Schedule a Demo →
#1 New AI Software Globally - G2 2026
Enterprise-grade security
Trusted by world-class engineering teams
