Capella hasn't kept up with AI and new MBSE capabilities; Is it time to abandon it?

Capella
1

There’s growing excitement about AI + MBSE, and a bunch of tools which lend themselves to AI.

But Capella wasn’t prepared for AI. So to me it’s been a long journey but it seems it’s finally time to abandon the project.

Alternatives

Recently I trialled a bunch of the latest tools:

  • SysIDE is AI enhanced VS Code add-in for SysML v2 code generation, with PlantUML model visualisation; alternatively all the big LLMs are pretty good already at SysML v2 generation.
  • Dalus.io is AI native and partially SysML v2 compliant, with the in-built LLM for generating system models, updating, querying, the model, generating, updating and verifying requirements all through a prompt interface where specific changes are reviewed and accepted by a human; Plus a clean Python interface for fully parameterised functions and behaviours.
  • Celedon Davinci is also AI native and fully SysML v2 compliant, for generating, updating, querying, and also word documentation.
  • Cameo Systems Modeler can be used alongside LLM, demonstrated by importing code from ChatGPT MBSE iNsights, or on the MBSE Execution channel which makes use of Model Context Protocol (MCP).

I’m still undecided, but it’s clear how well SysML v2 lends itself to LLMs and MCP. The System Engineer’s workflow is probably between 10x and 100x with these tools… querying, updating, drafting requirements, drafting architectures, even generating concept images all in the tool using prompts. The specialist engineer’s workflow too is boosted; like the Mech Eng rolling up the mass into a report in a few seconds, instead of making Excel mass budget workbook over the course of several days.

Using SysON was nowhere near as efficient; eventually I ended up using SysIDE to generate the model using code before importing to SysON as a diagram visualiser (not as clean as Tom Sawyer but good enough).

Where are we heading with MBSE?

MBSE Execution was eye-opening; seeing interaction between model and kinematic simulation, carrying out real-time verification of those requirements, running the animation and producing the report from a Claude conversation in like a minute, querying using the API, showing how the model integrates with DevOps (CI/CD) and other cool stuff.

Where are we heading with Capella?

What’s the state of affairs now? Nonplussed with the last answer ( Is Eclipse Capella going to use Generative AI? - #2 by StephaneLacrampe ) which was AI slop (once we understand the pattern of reaffirming opinions produced by AI). To me, Capella hasn’t kept up with the changing technologies, and with Obeo’s new pet project is SysON, Capella looks to fall to the wayside. AQL was always a no-go. I have zero chance of getting another living soul to use, learn about, or to debug AQL queries.

Capella is a big investment

The company training is not insignificant. The tool is free at face value but some of those paid add-ons are extremely expensive. From real quotes, I found Team4Capella is as much as 10x to 20x similar solutions in other tools, with a horrendously complex and messy deployment (not good for this little startup with no IT department). Next I went to Maple4MBSE, finding they are charging up to like $50,000 to $100,000 for a small company to link Capella to Excel.

We want to get the decision right if we’re going to continue to invest in an MBSE solution. Considering the sunk cost fallacy, I want to make the best decision now, going forwards.

Why are you investing in Capella going forwards?

I wonder if you might prefer the semantics, the readability and colour-coding, functional chains, the 5 viewpoints which everybody can understand (which is way better than all the SysMLv1/v2 diagrams, for everybody in the company except the SE) or some other reason like just the inertia of your company. How are you justifying the ongoing investment in a tool which doesn’t have these newer capabilities?

How do you justify moving forwards with Capella?

2

There’s a lot to unpack here, and the tone is a bit loaded, but the questions are legitimate. I’ll address them factually.

First Capella is not a drawing tool. It provides a proven structure to build, justify and evolve system architectures over time. It operationalize the Arcadia method which, in many organizations, is key to structure and share readability accross profiles (system engineers, management, domain specialists…)

Indeed Capella was designed before the LLM wave. But not “ai-native” (whatever that means, honestly…) does not mean “ai incompatible”. At Obeo we have ongoing initiatives combining Capella with LLMs and we are getting promising results. I’ve personally presented this work publicly, including a talk with demo at IEEE ISSE 2025 in Paris [1]. Our view is pragmatic, use AI to accelerate, discover, query, draft and automate while keeping what industrial engineering needs : control, traceability, human validation…
Among many findings we discovered that what makes Capella great to adopt for System Engineers, actually makes it also superior for LLM and agents compared to, for instance, using LLMs with SysMLv2 textual syntaxe.

Which brings me to another point. Capella and SysMLv2/SysON are complementary and we are investing in “the best of both”. We strongly believe Capella brings unique value, and we witness it every day with our customers. We also believe SysMLv2 and SysON bring major strenght: standardization, API-first workflows… This is not an “either/or”. The direction we’re pursing is explicitely about combining these strenghts. Capella and the desktop tooling is battle proven and has a rich eco-system. No doubt SysMLV2 will get a rich eco-system at some point if we keep going with SysON, but it’s just “a language”. We have great results in bringing the Capella and Arcadia vision into SysON. See [2] to have a glimpse of where we were last summer. Since then we kept working accross the whole stack (SysMLv2, SysON, Sirius Web, and the Arcadia Extension) and we start to have something exciting, the path is a just started.

“Team4Capella is 10x to 20x more expensive” , on this one I’ll just say that claim is just too broad to be meaningful and does not match our experience when comparing like-for-like scope. Comparing it to a lighter “file sharing” approach, or to a narrowly-scoped integration, is usually not an apples-to-apples comparison. If you want a serious discussion here, the only fair way is to compare the same scope: collaboration mode, security constraints, deployment model, support/SLA, audit requirements, etc.

So bottom line, to answer your question and speaking as Obeo : we justify moving forwards with Capella because it delivers durable, industrial-grade value for architecting complex systems: clarity, methodical structure, and long-program stability. And we invest in SysML v2/SysON because standards and API-first workflows matter. We believe the future is combining these strengths, not discarding one to chase the other.
We are actively working for that and Capella’s contribution will only be of higher value once we reach this combination.

[1] Keynote Speakers | IEEE ISSE 2025 | Paris, France
[2] https://blog.obeosoft.com/extending-syson-to-support-the-arcadia-method-a-first-experiment

3 Likes
3

A small contribution to the discussion: Capella’s key strength lies in its support for an engineering process such as ARCADIA, combined with a rich notation for describing complex systems. While Capella does not natively support LLMs, integration is still possible through the development of an add-on that leverages APIs provided by tools like Ollama.

4

What are you responding to? This looks like a nonsequitur.

I am very excited if AI is coming to Eclipse Capella; having been using the tool and following the progress since very early on. Though unless there’s a concrete plan to deploy as a fully capable feature in the next 6 months, I’d say it’s truly missed the boat if we are to be pragmatic.

The vision of Capella + SysON interoperability sounds great; I hope it works out, but I have to make the most pragmatic choices now to win and deliver projects this year and next.

Regarding expense, I’ll give an example, not broad, specific. Celadon Davinci core is ~€17/month/user; the capabilities, UI, and deployment is superior to Capella. Compare to T4C floating which was €373/user/month, or €13440/3users/year which works out to be the most expensive software in the engineering department, moreso than the CAD, FEA, Optics, embedded systems, multiphysics etc. and this is just ONE paid add-on, of which you’d need several big ones to approach the capability of some of these other tools.

The deployment of the trial took one of our software engineers days, riddled with problems. SysON took me a full day, changing UEFI settings and deep settings in Windows including getting Docker up and running. The deployment of the majority of the new wave of tools is near instantaneous, like Davinci, SpicySE, Dalus, fully web based, no docker, no software/IT specialists needed.

Capella was a leading tool at one time, but having looked what’s out there as of 2026, it seems it’s lagging and not gonna catch up.

5

I guess if your top priority is delivering this year with minimal setup effort for a small steam, fully hosted, near instant onboarding tools can be a pragmatic choice.

From our side we won’t turn a forum discussion into date-driven commitments, what I can say is:

  • Capella remains a stronc choice when you need durable architecture governance, shared readability accross roles, long program stability.
  • we have concrete Capella + LLM prototypes and we’ll keep sharing tangible work as it becomes ready to ship.
  • we are investing in SysMLv2 and SysON and on Arcadia support for those because standards and API-first workflows matter and we believe the future is combining strength rather than discarding one to chase the other. What makes Capella good now is still very relevant in this environment with AI agents.

On deploymeent friction: SysON is web-native, and will deliver something close to what you seem to expect, but it is not yet at the maturity level we’re targeting to enable a true “one click evaluation” experience. That gap is real and actively being worked on.

On more point that often gets lost in tool comparison : Capella is part of an Open-Source ecosystem. For many organizations, OSS is not an ideology, it’s a risk management and longevity choice.
Transparency, ability to audit and extend, reduced vendor lock-in, data privacy, and a community roadmap that is not solely dicated by short term priorities. That matters even more with AI because workflows and providers evolves quickly.

On pricing : headline numbers only become actionable when the scope is identical (collaboration model, security, operational constraints, support expectations, audit needs) and the needs are indeed highly contextual. Otherwise, readers risk drawing conclusions from non-equivalent products.

At this point, the trade-offs are clear: tool choice depends on constraints and goals. If you want to continue constructively, share 2–3 concrete use cases and constraints so the community can discuss options factually. Otherwise, I’ll stop here to keep the thread useful and avoid an endless back-and-forth.

6

I definitely agree with you. Indeed it’s possible, and can give good results, yet IMO one has to look beyond the excitement he can get by getting a few quick and shiny results and go further than that.

1 Like
7

I agree ARCADIA is a key differentiator, and probably the main reason I’ve been sticking around. The SysML + MagicGrid seems to have a steeper learning curve, and is so much more difficult for the specialist engineers to interpret. Getting buy-in is everything. Lagging so far behind the AI wave is now a deal-breaker.

If we had a production release of SysON/SysML v2 with the ARCADIA extension, along with the colour-coded elements, that would be worth sticking around for. A SysIDE style editor rather than the current SysON text importer would be a good step forward also. I’m sure an AI client for SysON will be on its way.

Paid add-ons are fine, but it’s important that there’s low-cost entry level solutions out there. I like how the new tools have free trials, then cheap plans with basic AI models and limited credits to get a feel for it, expensive plans for advanced AI models.