Spark is Campfire's spatial agent: an AI user that works alongside you inside Campfire, aware of the 3D models, documents, and spatial context in your scene.

In this demo, Campfire co-founders Jay Wright and Roy Ashok, along with Chief Customer Officer JJ, introduce Spark and walk through key capabilities. The session covers two full workflows: an engineering compliance review, where Spark cross-references a model against a spec document and logs findings on the spot; and a guided training procedure, where Spark leads a learner through a disassembly task, answers questions using scene and document context, and translates the instructions to Spanish on demand.

The full recording is below, followed by a complete transcript including Q&A.

Introduction

JAY

Welcome everybody. It is a pleasure to be with you today. We're excited to present where we're going with AI and show you a demonstration of Spark.

My name’s Jay Wright. I’m the Co-Founder and CEO of Campfire. I am joined by my colleagues, Roy Ashok, Co-Founder and Chief Operating Officer, and JJ, our Chief Customer Officer. I've had the pleasure of working with these gentlemen for a long period of time. We started working together at Qualcomm almost two decades ago, where we built a product and a business called Vuforia, that we built at Qualcomm and then subsequently drove at PTC.

So let me start off a little bit with our philosophy on AI. There's a lot that's happening in AI right now. There's a lot to be excited about. There's a lot to be confused about, quite frankly.

Our goal here is to use AI in a way that solves real problems. We're going to take the same approach that we did with XR and focus on the use cases where AI can deliver real business value and real use cases, and hopefully you'll see examples of that today.

Now the two things, the two problems, we're focused on the most and where we focused XR and AI is: one, making it easier for more types of people to participate in spatial workflows. So that means a wide variety of personas, whether they're engineering folks or people that have never touched a headset, or maybe even a computer before. And the other one is delivering it in a flexible way, so it can be used in a wide variety of workflows.

Now let me explain what I mean by spatial workflows. You might not have heard that term before. But you can think of that as all of the XR use cases that you've seen in the last 15 years. So in the past, we’ve talked a lot about design and engineering, training and education, sales and support. We have the same use cases today.

And the reason all these use cases have sustained is: they take advantage of one superpower that XR devices bring us. That superpower is the ability to fool our brains into thinking that digital content is physical stuff or physical places. And because it does that, it allows us to do work that previously required physical products in physical locations.

And of course, for those of you that have worked with this technology and implemented this in the past, you know the benefits can be staggering. It tends to vary by workflow, but, in general, brings both benefits above and below the line.

Now, while the use cases have not changed in the last 15 years, the tech absolutely has. So if you’ve caught up, or been up to speed, with devices, you know there's been a complete reset of the headset device landscape. We now have headsets that essentially deliver smartphone simplicity with app stores and amazing out-of-the-box experiences, so anybody can get up to speed very quickly.

AI — it's hard to overstate how fast AI has moved and is moving. But in, I think, less than two years, we've moved from it being a relatively thin layer of chat on an LLM to its own full application stack — and a huge amount of complexities come with it.

Now the other thing that has changed dramatically is software. Software for spatial workflows used to be custom applications for each workflow, tailored to a very specific device. And these would cost a lot of money to build, or buy, and oftentimes didn't scale very well. And some of you can probably attest to that, if you're sitting on applications on previous devices that might not be available, or vendors that might not be supporting those applications anymore.

Now, this is where Campfire fits.

Campfire overview

JAY

We describe Campfire as a spatial workflow platform, meaning a single platform that can deliver multiple workflows across multiple devices.

And I know many of you are familiar with Campfire already. So I'm going to start this demo with a very high-level overview of what Campfire does today, and then we're going to introduce Spark, our AI.

So, I like to describe Campfire as a combination of, maybe, PowerPoint and a first-person shooter.

Now, on the PowerPoint side, instead of creating slides with pictures and text — in Campfire, we're creating scenes with 3D models. And as many of you know, you can cut, copy, and paste 3D models to your heart's content in Campfire, and lay out everything from a manufacturing cell to a design review to a step-by-step procedure for work instructions.

And you can do it with a PowerPoint skill set. So, extremely easy to use. And if you haven't tried Campfire recently, I think you'll find there's a lot of new functionality that makes it easier than ever to create even richer scenes.

Now, where this really gets interesting is when we switch from the PowerPoint mode to the first-person shooter. So let's go into collaboration mode, just to give you a quick recap of where that stands. I'm going to come back to this scene with the engine and I'm going to invite JJ and Roy to join from an iPad and Galaxy XR, respectively.

Boys, want to go ahead and jump in?

Now, you'll notice here as they join, we see their icons up top, as you're used to. We see Roy as a rectangle. He's on an iPad, as we'd see other screen-based devices. And then we'll see JJ and his head appear in a moment. We see the head when we have any headset device. So there's JJ.

All right, so here we are. Here's our first-person shooter. And I'm not going to lie, we have lasers to play with here, also. So, we can go ahead and shine our lasers. I'm kind of having fun with shining and pointing the lasers, but as you know, this is an extremely powerful tool in Campfire, probably the one most commonly used.

There's also been an evolution of many of the tools in Campfire, with more powerful commenting capabilities as well as the ability to manipulate content.

So, it makes it very easy to collaborate on all these different types of workflows. Okay, there's our baseline. Let's go ahead and switch gears.

Meet Spark

JAY

Now, when we move to AI, you might think a lot changes in the user interface. It actually doesn't. And the reason not a lot changes in the interface is because we have delivered our AI as another user.

So we talk about Spark being embodied — maybe it should be em-headed, because he's actually just a head today, as opposed to a full body.

But you're able to interact with Spark just as you would another user, which means talking to him. And it's an interesting experience, when you're in there, you'll notice that Spark's head will point at you when you're speaking to him. And when you start working with content or pointing at things, you're going to see that Spark looks at that also. So it's aware of where you are. It's also aware of where the models are in the scene.

So I'm going to go ahead and switch views right now, so you can see exactly what this looks like on the headset. So, you can see Spark is looking down at the model right now.

JJ, why don't we just have Spark go ahead and introduce himself?

JJ

Spark, we're here doing a live demo. Could you introduce yourself?

SPARK

I'm Spark, Campfire’s spatial agent. I can answer all kinds of questions because I understand your 3D models, 2D documents, and spatial context within the scene. And I know how to use all the tools in Campfire, so you can ask me for help with anything.

JAY

Well, that's pretty good.

So, as Spark said, it's aware of the models inside Campfire. So, this scene has got this single robodog model. And as you know, when you add a model to Campfire, we also have the model and scene hierarchy along with it. So Spark has access to all that information in the hierarchy.

JJ, maybe we could just demonstrate that by pointing at a part and asking what it's called.

JJ

What is this called?

SPARK

That's the top cover.

JAY

Now, Spark's got the ability to answer questions using what's in the scene.

But we're also bringing together 2D data along with 3D models, into Campfire, available for Spark.

So that means you could take your existing documents, give them to Spark, and he'll have the ability to answer a question from those.

So we have created a synthetic manual for our favorite version of what this robodog should be. And we can ask some questions now of Spark. Maybe, JJ, let's ask what that top cover is for.

JJ

What is the top cover for?

SPARK

It protects the robot's internal electronics from dust, bumps, and bad decisions — basically its helmet, but with fewer stickers.

JAY

And if you look down to the left, you'll see Spark's response, where JJ can actually read the text that Spark has given. And you're going to see a button there that indicates a reference from the document that that came from. So you might see “robodogmanual.PDF”.

JJ, let's just go ahead and touch that and open it up.

So, in addition to being able to answer questions from the documents, you can bring those documents up manually. In fact, with Spark, if you just want to open a document and refer to that document by name, you'll now be able to do that in Campfire. So reference documents available to you, alongside models, and the ability to have Spark use them to answer questions.

All right, now, the other type of thing that Spark can do is it can actually invoke an action in response to a question.

So, JJ, I think we could demonstrate this by asking what robodogs are afraid of.

JJ

What's one thing that robodogs are afraid of?

SPARK

Drones. Even brave robodogs get a little jumpy when a drone starts buzzing around like it owns the place.

JAY

So in this case, this is functionality we have in Campfire today. You're familiar with hiding and showing models. Spark knows, when asked about that question, the response is to show that model.

That's also an example of Spark using Campfire tools. Spark can do hide and show. Spark can do any tool that you have in Campfire today. And Spark actually has access to tools not available in the UI, because they just might not make sense in the UI.

Maybe, just as another example, JJ, we could just lift robodog up 10 centimetres. Give it a little nudge up, get it more in our field of view here.

JJ

Raise robodog up 10 centimetres.

SPARK

Robodog is raised by 10 centimetres.

JAY

Those are some of the basics of what Spark can do. Now, where the magic happens is when you start putting these capabilities together into real workflows.

So, first, whatever workflows you're doing today in Campfire, Spark is going to make them easier and faster. And if for no other reason, because you can talk to Spark and not have to gesture or use a mouse and keyboard.

Whether you are presenting, whether you are creating new scenes for your content, or creating comments, Spark will deliver immediate utility.

But Spark also delivers new value and new types of workflows that become possible. And I'd like to show you a couple of those today. The first one will be an engineering review.

So, JJ, if you want to go ahead and let's move to our engineering scene, and we'll talk a little bit about the setup.

JJ

Next scene.

SPARK

Moving on to Acme Company.

Engineering review example

JAY

Now, one of the most common engineering use cases is reviewing a particular product or machinery or process, and making sure that it is compliant with a range of different specifications.

And today, the way that's happening in Campfire is: you put together your Campfire scenes, you might have one or more person that jumps in with a headset, other people on other devices, and you've got the person with the headset sort of validating or exercising those requirements.

But the requirements are not in Campfire. Those requirements happen to be in a separate document somewhere else. Well, now that Spark has access to the models, and awareness of the models in the scene, and awareness of the documents — Spark can help you find those issues very quickly.

So, JJ, I'm just going to walk up to the cell with the observation that that door is too narrow. Can we have Spark just check that against spec?

JJ

What's the width of this door; does it meet our spec?

SPARK

The door is 80 cm wide. The spec calls for 100 cm minimum clear opening. So, it does not meet the requirement.

JAY

So again, this is no training data provided. All we did for Spark to be able to do this was give it the models, as you do today, and give it a document — which in this case was PDF — and Spark was able to make that comparison for us.

Now, it's great. It helped us find that very quickly, looked up the document. But what we really want to do is capture that issue right now, so that we don't lose it for later. And I think the way a lot of folks are doing this today is kind of capturing those issues or comments in a different place — because I think we've all experienced the challenges of text entry on VR headsets, and it's not amazing.

JJ, let's go ahead and document that.

JJ

Can you add a comment to that effect?

SPARK

Added a comment noting it is 80 centimetres, and below the 100-centimetre minimum.

JAY

So, just like that. Something that might have taken 15 minutes, an hour, maybe not even been able to collect all that information at one time — we got it immediately, we documented it comprehensively, and now it's in a Campfire comment that can be sent with a link or shared elsewhere.

JJ, I'm going to go ahead and make another observation, with my safety hat on. I feel like there ought to be placards here. Are there placards missing, safety placards missing?

JJ

Spark, I don't see any safety info. Should we have a placard here?

SPARK

Yes, you should. Post warning labels on the approach side of each access door that say “warning” and “authorized personnel only.”

JAY

I think we ought to add that one as a comment too.

JJ

Can you add that here as a comment?

SPARK

Comment added with the required warning text for this access door.

JAY

Simple. Straightforward. Issue captured.

All right, that was great. Let's go ahead and move on to another use case, more in training.

Guided training example

JAY

Now, I know many of you are familiar with using Campfire for training today. It's used widely for instructor-led training, where you can put together your curriculum and share it and have an instructor run other people through it.

But what about on-demand training? What we can do with Spark is actually turn Spark into our trainer.

JJ, let's go back and see our friend the robodog, and take a look at what that training experience can look like.

JJ

Next scene.

JAY

All right, there's our friend the dog. Now we're going to do a very basic procedure, a three-step procedure that Spark will walk us through, to replace a battery.

JJ, let's go ahead and have Spark kick this off.

JJ

Guide me.

SPARK

Hi, I'm Spark, your trainer. Today I'm going to show you how to replace the battery pack. Any questions before we get started?

JAY

I have zero questions.

JJ

No questions. Let's get started.

SPARK

The first step is to remove the top cover. You will need a 3/8-inch hex driver. Let me walk through this step with you. Remove the screws and the cover will lift right off.

Let me know when you’re done.

JAY

So you can see, we’re starting to put together all the pieces we've seen before to deliver this new workflow.

Spark is reading out instructions for this, telling the user what this step is for. Spark is then performing an action, which in this case is an animation — which, by the way, Spark helped us create. And now Spark is waiting for our learner to ask questions.

So, JJ, I'm looking at this and I'm pretending I’m holding a screwdriver there, and I want to make sure I unscrew all those. I want to probably count as I do it. So can we just ask Spark, like, how many screws there are? And then I'm just going to go crazy and count unscrewing them.

JJ

How many screws are there?

SPARK

Let me think this through for you. There are 16 screws securing the top cover.

JAY

So again, this is a question that was answered just on the information that was delivered in the scene and model hierarchy. We didn't have to do anything new to make that accessible.

All right, I think I'm done. I just unscrewed my screws really fast.

JJ

I'm done.

SPARK

The next step is to pull off the front cover. Use both hands to carefully lift, and pull the front cover toward you.

Let me know when you're done.

JAY

You saw that step sort of had the lift and pull. I'm going to go ahead and guess that there's some people, myself included, that would just start pulling on that thing, to the point where it feels like it's going to break. And I would want to ask what I did wrong. It's pretty hard to just ask a user manual, but if I have Spark, that's got access to other information, I might be able to get a tip.

So, JJ, can we just tell Spark I'm having a hard time pulling this thing off. What might I be doing wrong?

JJ

Front cover is hard to remove. What do I need to do?

SPARK

Use both hands to lift first, then pull the front cover toward you as shown. Don't force it. If it still feels stuck, stop and avoid excessive force. I can highlight the front cover to confirm the target part if you want.

JAY

Well, that was a nice offer. Sure.

JJ

Highlight the front cover.

SPARK

Front cover is highlighted so you can confirm the right part.

JAY

Again, Spark is using tools to provide answers visually. But now we're creating an instructional experience that, in some ways, is even better than having a human guiding through it.

JJ, let's go ahead and move to our final step in this complex procedure.

JJ

I'm done.

SPARK

You're almost done. The last step is to remove the battery pack. Pull it gently with even force to slide it out.

That's it. You've just removed the battery pack successfully. I'll send the recycling instructions via email.

JAY

So one of the challenges of creating curriculum at scale is dealing with language and localizing, and the power of AI sort of takes care of that for us.

So, JJ, let's go ahead and just have Spark repeat that step in Spanish.

JJ

Repeat that last step in Spanish.

SPARK

[Spark repeats the battery removal step in Spanish.]

JAY

Excellent. So, as you can see here, Spark is able to deliver on-demand workflows in a way that was not possible before. And is able to do that, and you're able to create this, very simply.

It's a matter of giving Spark basic instructions for each step, documents that it can use to answer questions, and then creating animations, if you want Spark to be able to animate and move things, as we did with the screwdriver turning and removing the screws.

So we really think this is taking guided instruction to a whole new level, closer to having an actual person or assistant than something that has been traditional interactive content.

Spatial intelligence + what's next

JAY

But there's another dimension here. And for those of you that are using AI every day and rely on memory, you know, AI has memory of the transcripts of our conversations with AI.

Now, that same kind of concept applies in Campfire, but it's beyond just the conversation between you and Spark.

We call it a spatial transcript. And what Spark is recording and collecting is not only what's said, but also the interactions — meaning what you're pointing at, what you're moving, the commands you give, and even how the model is moved, if people pick things up and move them with tools.

Now we see this new data, this new layer, as a new form of spatial intelligence that just comes with using Spark and using Campfire. And we see this data leading to valuable insights that were not possible before.

I mean, imagine asking questions about — what are the most common issues we're finding in such-and-such a program, at such-and-such a location? Where are people getting stuck on the new training for XYZ product or process?

All those questions, all those insights, become possible through that spatial intelligence layer.

So we could not be more excited about what Spark's going to deliver for what you're already doing today — just making existing workflows much more seamless, much more powerful, much more valuable.

But then also, the notion of guided workflow, where you can take people through content. And I should point out also — we showed a guided example here in a training scenario, but guided instruction goes beyond training.

For example, let's say we wanted to create a simple product explainer where I present my product, and I either want a customer or maybe a supplier to walk through it. I want Spark to explain certain attributes, and allow my customer or supplier to ask questions. You can do that extremely easily now.

You can put your models in Campfire, you can give Spark some basic instructions, share that workflow — and back comes the results, or the questions and answers that your customer had. So, just a completely new type of guided workflow experience that becomes possible.

Now, in terms of availability for Spark, we are hard at work on Spark for a preview later this year. The best way to get access and stay in the loop is go to the Spark page on our site and you'll see a waitlist there.

Now, we're engaging with partners on a rolling basis, depending on use case. So the use case that is front and center is guided instruction. So, those of you that might be looking for a path for HoloLens and Guides-like experiences, please jump in. We'd love to engage. We want to make sure that Spark meets those requirements and absolutely takes what you're doing to the next level.

Q&A

JAY

All right. I think that covers everything we've prepared. Meag, maybe we open up for some Q and A?

MEAG

Yeah, we've got some questions waiting. Jay, there's a question asked when the drone appeared: Can Spark only show models already in the scene, or can it pull models from my library on the fly?

JAY

That's a great question. Spark can definitely pull from the library on the fly during the authoring experience. So if you are creating scenes as an author, you can just say, add the drone from the library. Absolutely.

MEAG

We had a question — were documents loaded in Spanish, or was that an automatic feature built into Spark, when we did the Spanish instructions?

JAY

Yes, Spark was translating from English documents. So if you don't have translations, you can put what you need in there in English, and let Spark do the rest.

MEAG

What LLM powers Spark? Can we bring our own LLM, cloud or offline local?

JAY

Spark's being developed on OpenAI. With regard to using your own — yes, you can use your own, but today it needs to be OpenAI. The way that that's implemented is with our hybrid cloud architecture.

With hybrid cloud today, all of your model data is living in your own Azure tenant. Spark with hybrid cloud means Spark's memory and all those documents would be within your own OpenAI instance, in your own Azure tenant.

MEAG

We just had a question come in as you were explaining that, about proprietary projects and how can we be sure that the information stays safe and is not shared publicly? So you just touched on hybrid cloud, but I think, maybe we could answer that one for both managed and hybrid.

JAY

Yeah. To be very, very clear — our hybrid cloud option has been designed so that the customer has control of all data and we do not physically have access to it. So today that holds true for all model data that you're putting inside Campfire. It already works that way. We're using that same concept with Spark and AI, which means we can't see it, and we certainly can't use it for training or any other data.

We are giving you the tools to create and build your own spatial intelligence layer, for your benefit.

MEAG

We have a question here. Someone who is an EHS manager, wondering about how they can use Campfire. Specifically they asked, could we use it to run spatial job hazard analysis reviews, walking workers through hazard zones directly on a 3D model of our workstation?

And also, we need to plan and validate overhead crane and hoist operations for lifting aircraft engines. Can we simulate lift paths and clearance envelopes in 3D?

JAY

Yeah, I mean, that's exactly the type of activity that happens with Campfire in an instructor-led scenario today. So I think you should be able to do all of those things.

I think the things that I'd probably want to pull on would be the types of simulations that are required for hoisting. There's animation capabilities that Spark has and can use automatically, but I think that should all be doable.

MEAG

A question here. Does Campfire support execution and animation of work instructions: assemble, disassemble, highlight each steps/parts/tools/kits, and would Spark follow users executing procedures and correct their actions if they make errors? And is model or area tracking available? Those are kind of all one set.

JAY

Yeah. Yeah, sure. So, I think, on the second part of the question, we're starting off with Spark delivering guided instruction, but not the ability to verify that a user has completed it. That will come, but we're taking this one step at a time.

Can it do work instructions? One hundred percent. Yeah. I think, if you're familiar with Guides, and you have done something with Guides, you're probably going to be elated with what you can do with Campfire.

But yes, you can create multiple steps in a procedure. You can separate your steps from your scenes. So if I had a procedure where I wanted Spark to walk through three different steps on a scene and highlight and animate three different things in sequence and provide written instructions and documents alongside that step — all that is possible.

MEAG

Sorry, I just read that last part of that question. It said, is model or area tracking available to position the 3D models on products or in spaces? So that sounds like scene anchors to me.

JAY

Yeah, interestingly enough, you might think since we spent decades of our career building model tracking technology — we don’t, we don't have model tracking in Campfire today. And I'll tell you, there's a reason for it. I think we found that one of the key learnings is it really takes a long time to set that up and configure curriculum and content with different models. And while it's amazing when it works, it's not something that we felt was sufficiently robust and sufficiently scalable, to start with. That's not to say it will not make its way into the product, but that is not in there today.

There is a way to achieve registered content, though, and that's through the use of a scene anchor. So we have taken the tried-and-true approach of a fixed fiducial that you place in your environment. So if you want to put this fixed fiducial QR-code-looking thing next to your equipment, you can achieve that same effect in a scalable way.

MEAG

I have a question about hand gestures, for speaking to Spark. Can we turn them off on all headsets, except only have one headset that has access to hand gestures?

JAY

The way Campfire works is, when somebody shares a workflow with you, you're assigned a certain role. And in a situation where you're doing training, that role is probably going to be a viewer, so they can just kind of click through and view, or a commenter, where they can leave comments.

So in the situation where somebody's a viewer, you're always going to be able to see their hands, and they can see their hands and point, but it's not like they'll have the ability to use gestures that are going to change anything in the scenes or mess anything up.

MEAG

We do have another question. Can we display live data, such as on-machine sensor values (eg, operating temp)?

JAY

Not today. Not today. Love it. We'll get there, but not today.

MEAG

I have another question. Is Spark only available on headsets?

JAY

Yeah, this is a good one. We demonstrated Spark on the headset and showed that experience today. Spark also works just as well, and frankly equally compelling, on screens. So, you can speak to Spark like we did on the desktop, but there's also a chat-based interface on the desktop.

It's probably also worth noting, on the device side, that we're expanding headset support with guided instruction, to pick up the new Project Aura device. So again, I think those of you that are looking at a path for a HoloLens and Guides experience — we think Campfire paired with that device is going to be a dynamite combination.

MEAG

There's one here. I almost missed it. I don't know if this is actually specific to Spark, but on the cell review scene, someone asked: can you animate that robot according to a KRC (Kuka) motion file?

JAY

We can't take the motion file specifically, but you should be able to export that as an animation, into GLB, and bring that into Campfire, so that you will have the animation that reflects robot programming in the Campfire scene.

MEAG

Okay, I think that's all of our questions there, Jay, if you want to put a bow on it.

JAY

Excellent. Well, thank you again for joining. We're very excited about where this is going. We're very excited to engage with all of you to understand the workflows where you'd like to apply this. So please jump in, get on this waitlist, and please share what you can about the use case that you're focused on so that we can reach out and connect.