Adam CAD Copilot

The AI CAD co-pilot living inside Onshape and Fusion uses natural language to complete component editing, feature tree cleaning and parametric reconstruction.

design CAD Mechanical design

https://adamcad.com/

Visit Website

In-depth Report

Adam CAD Copilot is an AI CAD auxiliary tool that runs inside Onshape and Autodesk Fusion, allowing mechanical engineers to use natural language to complete component editing, feature tree cleaning, and parametric reconstruction, instead of starting from scratch to generate an unusable concept drawing. It appeared on the Product Hunt daily list on July 1, 2026, entering the top five of the day. The company behind it, Adam, disclosed that it had completed approximately US$4.1 million in seed round financing according to Chinese media. Its core selling point is very restrained: do Copilot first, not Autopilot, all changes remain editable and traceable, and important actions are returned to manual approval.
Adam’s company positions itself as an “AI workspace for hardware teams,” and CAD Copilot is just its entry point for engineers. An earlier public action came from Onshape: Onshape’s official blog announced in March 2026 that Adam AI would be available on the Onshape App Store as a free open test (open beta), which will be used for feature tree optimization within Part Studio. On July 1st, Adam logged into Product Hunt with a more complete "CAD Copilot" form, focusing on "AI CAD inside Onshape and Fusion", and entered the top five in the daily rankings that day. In terms of financing, the public account BIMCorner mentioned in its industry review in March 2026 that "AI native CAD platform Adam completed US$4.1 million in financing." However, this figure has only been mentioned on the side in Chinese media. The official has not disclosed the complete round and investors on the landing page, and the company's subsequent official announcement should prevail. In terms of positioning, Adam is obviously betting on a judgment: "engineer-specific tools" such as CAD, EDA, and PCB will be rewritten by AI, but it chooses to start from assisted design (Copilot) rather than directly replace design.
The official CAD Copilot page lists four core workflows. The first is prompt word-driven component editing: replace the original bunch of tiny clicks with one sentence of instructions, such as "Add the wall thickness from 2 mm to 3 mm" and "Add four mounting holes with a diameter of 5 mm." The second is selection context: using the selected face, edge or entity as the context, allowing instructions to accurately fall on the correct part of the model, rather than describing them out of thin air in a dialog box. The third is feature tree optimization: automatically find duplicate or redundant features and merge and rename them to make the model more robust and easier to understand. The fourth is parameterization and variable extraction: extracting hardcoded dimensions into named variables and cascading references throughout the design. The biggest difference of this product is that it does not convert images into 3D. General-purpose AI 3D generators tend to spit out a static rendering that is divorced from the engineering context, while Adam reads the existing feature tree and geometry, respects the existing parent-child reference and parameter relationships, and returns a CAD feature that is still editable. Onshape's official description also confirms this: Adam renames features, merges redundant operations, exposes hard-coded values into named variables, and can also change basic edits such as fillet radius in Part Studio. The official FAQ also emphasizes that when an edit will destroy the father-son relationship, Adam will not forcibly rebuild it, but will mark the risk and return it to human judgment. If you zoom in to Adam's larger workbench, it can also connect to Onshape, Fusion, SolidWorks, Arena (PLM), McMaster-Carr, Digi-Key, Xometry, etc., receive instructions through Slack or email, directly output branch model editing, BOM after reconciliation, updated drawings, renderings, draft RFQ inquiry orders, and link back to the source. In other words, Copilot solves the step of "modifying the model", while Adam's main product takes care of BOM, source finding, rendering, and supplier communication after modifying the model.
The pricing of the Adam main site is divided into four tiers: Free (including 300 daily refresh quota, shared folders, custom models, persistent memory, multi-modal generation, Slack/email usage, 3000+ integrations); Starter is $40 per month (4000 quota + Ultra model); Pro is $200 per month (20000 quota + cloud storage, marked as "most popular"); Max is $1000 per month USD (100,000 limit). The copilot page is labeled Free Start on Product Hunt. There is a point that needs to be reminded here: Product Hunt marks Copilot as Free, but the paid version of Adam’s main website is for the entire Adam product, and the public materials do not clearly separate Copilot’s exclusive quota, restrictions, and support terms from the large package. ToolWorthy's review also pointed out that Copilot's specific usage limit is not clearly separated from Adam's broader Free/Starter/Pro/Max pricing, and teams should confirm applicable terms directly with the official before large-scale adoption.
Since the product is in its early stages, there are not many public reviews from real users, and more are the reviewers’ inferences based on official demonstrations. The positive judgments given by ToolWorthy focus on several points: it is highly consistent with the real workflow of mechanical engineers, rather than a generalized AI efficiency tool; it lives directly in Onshape/Fusion, which engineers use every day, and does not require migration; it focuses on editable CAD structures rather than one-time visual output; feature tree optimization and parameterization really hit the pain points of engineering. AIPURE's Chinese evaluation also agrees that selection-aware prompts can be more accurate and faster than plain text interaction, and feature tree cleaning and parameterization improve model maintainability. Negatives and reservations are equally clear. First, it is essentially a tool for "modifying/cleaning existing CAD". If what you need is to model from scratch in an unsupported tool, its value is limited. Second, performance is highly dependent on the quality of the underlying feature trees and constraints—an inherently messy model may require multiple rounds of bootstrapping. Third, reviews generally believe that the generated edits are still engineering work and must undergo manual review. chatgate.ai clearly recommends: before releasing the official part, first test it with a copy of non-critical parts to confirm that the changes retain constraints, downstream references, configurations, drawings, tolerances and manufacturing feasibility.
Onshape's official blog is a relatively independent first-hand information source. It defines Adam as "AI collaborative driving that optimizes feature trees within Part Studio." The focus is not on generating new geometry, but on making existing models clearer, more parameterized, and easier to maintain. This positioning is consistent with Adam's own caliber, and also explains why it can gain a foothold in the Onshape App Store - cloud native architecture plus API driver, allowing third-party applications to directly operate on Part Studio. Reviewers generally put Adam into the narrative of "AI is replaying the AI programming script": in 2023, Copilot changed to writing code, and in 2026, similar copilots began to enter CAD. chatgate.ai believes that Adam is most suitable for teams that have repeatedly iterated in Onshape/Fusion and frequently cleaned models; but the value depends on whether you actually use these two CADs and how much time you spend on repeated editing and feature cleaning. The comments on the Chinese public account "Node Zero" are very accurate: Adam's entry point is pragmatic - do Copilot first, not Autopilot.
We haven’t seen any major controversies or security scandals surrounding Adam, but there are several types of risks that engineering teams should evaluate before plugging in. One is the issue of data ownership and training: the official FAQ promises that your geometry, drawings, BOM and transaction records will be kept in your own account and PLM. Adam only reads the content required for the task through limited authentication access (try to use OAuth), does not copy your CAD to independent storage, and does not use any connected data to train the model. The workspaces are isolated from each other and encrypted; the enterprise version adds SSO, audit logs and administrator permission control. Whether such commitments are valid should be confirmed by retention, governance and security reviews during the company's onboarding. The second is that the business terms are not transparent: Copilot’s free label is not equivalent to the full pricing model, and the quota, privacy and support details are unclear in the public materials. The third is product maturity: as an early product, integration stability and reliability should be tested on real models instead of just watching demonstrations. The fourth is engineering responsibility: any automated edit cannot eliminate manual verification of constraints, tolerances, and manufacturing feasibility.
Adam CAD Copilot is most suitable for three types of people: mechanical engineers who do iterative part design in Onshape or Fusion; hardware teams who want to make feature tree cleaning and parameterization a regular job; and startups preparing CAD files for design review, sourcing, or manufacturing handover. It is not suitable for teams looking for a stand-alone conceptual model generator, nor for organizations that primarily use other CAD platforms and struggle to standardize. The usage advice is straightforward: try it out on a real but non-critical part first, verify that the output remains editable, and then check whether the feature tree cleanup makes downstream changes smoother or messier. Pick a model with a messy feature tree that you can understand, and have Adam rename the feature, merge a safe duplicate, expose two dimensions as variables, do a selection-aware edit, compare the reconstructed model, and run through the team's usual drawing, interference, and fabrication checks. After confirming that the history is clear and subsequent parameter behavior is predictable, the authority can be gradually expanded.
The value of Adam CAD Copilot does not lie in "using AI to draw a new thing", but in that it lives faithfully in the engineer's existing Onshape/Fusion, turning the most labor-intensive repetitive editing and feature tree cleaning into a natural language command, and always retains the editable, traceable, and approveable project background. It is still an early product, and its business terms and maturity need to be tested, but for hardware teams that rely heavily on these two CADs, it is likely to be one of the first "CAD co-pilots" that can truly be implemented.

User Reviews

EMjac

—

The open source version can be easily deployed on GitHub, and the commercial version can be copiloted in Onshape and Fusion to collect money. This dual approach of open source and commercial is much smarter than pure closed source.

Katherine664

—

I came across CADAM in Qiniuyun's weekly report. It is a text-to-CAD web application open sourced by the Adam team. To put it bluntly, it is typing to generate a parametric CAD model. I went to GitHub to check the warehouse. Local deployment is not difficult. The modeling threshold for 3D printing players has really been cut off by AI this time. The group of people who were discouraged from learning OpenSCAD can come back.

DMorganQ16

—

The good news for 3D printing parties is that you can produce models just by typing.

TWatson

—

When we work in consumer electronics, we often have to sort out a bunch of messy feature trees before design review. In the past, senior engineers spent half a day manually renaming and merging redundant features. Now they are left to Adam to publish the first version, and people scan it again. The time saved is not particularly large, but the advantage is that you don’t have to start from scratch every time. The maintainability of the model is visibly improved, and newcomers who take over the old model are no longer confused.

Benjamin.Morgan_2022

—

We use Fusion. After installing the AdamFusion plug-in, we run "parametrize my model". It extracts the hard-coded dimensions into named variables, and then changes a value to the whole tree cascade. It is much faster than manually finding each dimension. However, I will still run through the team's drawings, interference and manufacturing inspections before I dare to use it. After all, the generated things must be manually reviewed and cannot be fully trusted. Engineering responsibility cannot be escaped.

悠然130

—

To be honest, I refused to modify CAD with AI at first, for fear that it would collapse the feature tree. Later, as suggested, I first took a copy of the non-critical part and tried it, allowing it to rename features and merge a duplicate. After comparing the reconstructed model, I found that the history was indeed clear and the downstream parameters were not messed up. Now the old models that are cleaned and imported are basically handed over to it.

RonaldBennettZ

—

What strikes me most is that it respects parent-child references: when an edit would break a dependency, Adam doesn't force a rebuild, but flags the risk back to manual work. This design philosophy of "cautious entrustment" is more reassuring than "fully automatic modification for you" and is more in line with the real constraints of the project site.

JackSanchez_X14

—

There is no going back.

邹月玉

—

For our team working on hardware, the most annoying thing is BOM misalignment. Adam can automatically mark the parts whose versions do not match, and find the ones that are still in WIP among the 47 materials, which is really worry-free.

JHill_88

—

Feature tree savior! !

aRTHURcOX

—

The biggest difference between it and the general AI 3D generator is that the result is still editable and conforms to engineering logic, instead of throwing you a concept map out of context. For a team like ours that needs to be used for proofing, drawings, and ECO, this is the key.

txjid3rk3

—

It's so cool to change the model directly in Onshape, and finally I don't have to cut it back and forth.

GeorgeStephens_Pro

—

Cursor for CAD, absolutely.

Margaret_CampbellII

—

The Pro version costs 200 dollars a month, which is a bit of a pain for a small team.

Zachary_Evans_2023

—

The team that came out of YC Winter 2025 has US$4.1 million in financing behind it. It can be seen that they want to do CAD seriously, not just for fun.

sadtiger241

—

I tried it out and asked Adam to increase the wall thickness from 2 mm to 3 mm. The changes are editable and the history is still there. This is much more reliable than those tools that generate a mesh.

JMurphy_Pro_530

—

Pricing is a bit confusing. Product Hunt marks Copilot as Free, but the main website has four tiers: Starter 40 dollars, Pro 200 dollars, and Max 1000 dollars. It’s not clear how much Copilot accounts for and whether there are any separate restrictions. ToolWorthy's review also specifically pointed out this pitfall, saying that the public materials did not separate the Copilot dosage from the large package. We are preparing to roll out with a small team. Before scaling up, we must clarify the terms with the officials before making any move.

Anthony.EvansK

—

The data is not used to train models and is kept in their own PLM. This is very important to corporate customers, otherwise no one would dare to connect the drawings.

DavidMorris_Pro

—

After reading founder Zach’s original words on HN, they decided from the beginning that “CAD as code” was the right abstraction, using Onshape’s FeatureScript to directly interact with Python in Fusion. This idea is much more advanced than simply setting up a chat box to generate STL, and it is more like a tool for engineers.

091dy3ffj

—

The renderings are also pretty cool, with hero shots and explosions coming out in one sentence.

KeithKristensen

—

Only Onshape and Fusion are supported. Our company mainly uses SolidWorks, so we won’t be able to enjoy this wave of dividends for the time being.

Parker515

—

Has anyone tried using it on the official release? Or do you only dare to play on copies?

PEpar

—

Someone on HN asked why not just use open source like OpenSCAD. An engineer immediately replied: OpenSCAD doesn’t do Brep, and professional teams won’t use it. This makes sense. Adam's bet is originally on the workflow of existing commercial CAD.

Jack_PerryQ203

—

Free ones go first.

Adam CAD Copilot

In-depth Report

User Reviews

Similar Products