Episode 37

AI-Native Toolchains with Thomas Froment - Eclipse Foundation

Thomas Froment, Program Manager for Development Tools at the Eclipse Foundation, talks about how AI-native, vendor-neutral tooling is transforming the way IoT and LoRaWAN developers build, test, and ship products.

In this episode, he explains what Eclipse Theia is, why it matters, and how open-source toolchains give companies more control, privacy, and long-term resilience than proprietary AI editors. Drawing from his experience leading Theia, Open VSX, and other Eclipse development-tool initiatives, Thomas breaks down the rapidly evolving AI workflow landscape and why embedded engineers should pay attention.

  • What Eclipse Theia actually is: a framework for building fully customizable, AI-native development environments designed for embedded and IoT toolchains
  • How Theia differs from VS Code and Cursor, including privacy, extensibility, transparency, and the ability to integrate hardware, local workflows, and cloud systems in a single toolchain
  • Why open-source governance and vendor independence matter for companies developing IoT devices, especially in regulated or security-sensitive environments
  • The explosive growth of Open VSX and the shift toward extension ecosystems not controlled by a single vendor
  • The role of Model Context Protocol, AI agents, and domain-specific prompting as organizations integrate AI deeply into engineering and testing workflows
  • How teams use Theia to build hybrid local-plus-cloud development environments that support hardware-in-the-loop testing, device constraints, and long-tail IoT edge cases
  • Emerging use cases for lightweight and local AI models inside IoT products, and why customization of prompts and agent behavior becomes essential
  • Collaboration tooling within the Theia ecosystem, enabling real-time co-editing, code reviews, and multi-developer workflows for embedded teams
  • Why IoT and LoRaWAN companies need to think in terms of entire toolchains rather than just IDEs, and how open-source components allow a tailored pipeline from development through testing and deployment

Guest Links:


Transcript
Speaker:

My guest today.,Thomas Froment,

is not a LoRaWAN native.

2

:

He's the program manager over at Eclipse

Foundation, running the Theia program

3

:

and walking people through

how AI is going to be integrated

4

:

into businesses and specifically open

source tool chains like Theia.

5

:

So we talk about the difference

between Theia and Cursor,

6

:

what Theia can do, what it might be good

for if you're looking for privacy

7

:

or customization or security versus

using some of the other tools out there.

8

:

Fascinating conversation

with a guy deep into AI

9

:

and really relevant to what we are doing

in the general IoT world.

10

:

This show is sponsored

by the Helium Foundation.

11

:

To check out what they're doing.

12

:

Go to Helium.Foundation.

13

:

Now let's dig into the conversation

with Thomas Froment.

14

:

Thomas, welcome to the show.

15

:

Thanks so much for coming on.

16

:

So thank you for inviting me.

17

:

I'm really excited

to have you on as kind of this

18

:

not outsider to LoRaWAN,

but certainly not this LoRaWAN core piece

19

:

is that you're coming into this

from an AI core piece.

20

:

So let's probably start off with what

Theia is so people understand

21

:

what we're talking about

before we get into it. Yeah. That's true.

22

:

Thank you

I was going to start by a disclaimer

23

:

saying, Okay,

I am not not a LoRaWAN expert

24

:

I am working for the Eclipse Foundation

as a program manager.

25

:

I'm working on the development tools

part of the Eclipse Foundation.

26

:

And as you said,

Theia is one of the key project

27

:

and AI capabilities of Theia

Theia’s platform is a framework

28

:

to develop tools

and to develop IDE like tools.

29

:

It's a kind of new generation

of tooling as the Eclipse Foundation,

30

:

and it comes with a very exciting

and very great AI features.

31

:

And like all the open source project

at the Eclipse Foundation under

32

:

under no governance, so hopefully

we will come back to this later on.

33

:

But yeah,

34

:

this is also why I accepted to join you

in this podcast because yeah, I say, okay,

35

:

we have people who have more expertise

on any on this topic,

36

:

but we know many companies were already

using this technology to develop

37

:

some tooling and building blocks

or to customize toolchains.

38

:

And why Theai?

39

:

I don't know if you want me to,

to go into all the details.

40

:

Right now,

but it's an open source project.

41

:

So we'll open source framework

with building blocks and fronts.

42

:

To give you a few example,

when you start doing AI,

43

:

you want some basic capabilities

like connecting to what we call MCP.

44

:

MCP, the kind of standard way

of connecting to any system we have.

45

:

You have thousands of systems

46

:

if you want to connect to any tool,

a check system,

47

:

a corporate system, or whatever

your drive, Google Drive, Google White,

48

:

whatever you want and you want to cut

when you want to connect to this stuff.

49

:

Of course,

you need some software components.

50

:

And in the Theia open source project

you have this component

51

:

you can integrate then and you can

make them part of your own toolchain.

52

:

And exactly why I say, okay, it's

very relevant

53

:

to show what you have in

as part of this framework.

54

:

I just give the model Context protocol,

the MCP,

55

:

as an example of what is available

among the building block of tier.

56

:

But you have basically all the basic,

most important block.

57

:

Of course, to drop agent as well.

58

:

Literally today,

more and more people are looking at

59

:

how to integrate

agent AI agents into their systems.

60

:

And for

61

:

that's again, yeah, it's

if you start from scratch

62

:

or you try to develop it by yourself,

it's yeah, you will you will experiment.

63

:

And basically today with all this tool

or it's moving very, very quickly.

64

:

We have a lot of innovation.

65

:

And every day even myself,

I discover a new tool

66

:

and sometimes a new tool having

after a few weeks, millions of users.

67

:

So people are really excited

about using this technology

68

:

because it brings great benefits

in terms of productivity and so on.

69

:

But it comes also comes with new risk.

70

:

You guys are seeing usage grow yourself

right that you're seeing this in Theia.

71

:

Yeah we see usage grow.

72

:

This is one of the

you know it's a clear foundation.

73

:

Most most of the time

people know the foundation.

74

:

Thanks to the

the eclipse ID which is 20 years old.

75

:

And but at the foundation

we have many new project.

76

:

And this one here is one of the most fast

growing project we have.

77

:

We are releasing, a release every.

78

:

When I say we, it's not myself,

it's a community.

79

:

It's a community based core of the open

source developers are delivering

80

:

a new release every two weeks

with, more than 100 contributors.

81

:

So you see, it's a very,

very active project,

82

:

and they are basically integrating

the most recent AI features.

83

:

People are working on new models,

84

:

new start outs, protocols

that are coming all the time.

85

:

So yeah, it's

one of the most fast growing.

86

:

And along with tier

we have the marketplace called Open Race.

87

:

It's also a part of my very Mr.

88

:

Foundation

as a as working on the development tools.

89

:

It's another open source project.

90

:

But it's not only a project,

it's also a service.

91

:

And basically it's an alternative

to the Microsoft marketplace of extension.

92

:

I don't know if people are familiar with

this system, but basically the VS code

93

:

extension mechanism is a kind of defacto

standard for extensibility

94

:

mechanism and software

people are using VS code.

95

:

Was attending

96

:

the things conference recently

and basically everybody was using VS code.

97

:

And we know that

and as extension mechanism,

98

:

we are relying on a marketplace.

99

:

And one of the key feature of to to be

compatible with the standard marketplace.

100

:

So you can connect to as an IDE

or as a tool using building blocks.

101

:

You can connect it and benefits

from the Microsoft bucket of extension,

102

:

or you can use you a separate marketplace,

the VS code.

103

:

Open this one.

104

:

And in that case with tier

plus openly six,

105

:

you have a fully independent and not no.

106

:

You know you cut any risk to be one door

locked would say

107

:

or to have any dependency

on a single though.

108

:

And this is kind of the big thing here

is that a lot of folks are using VS code

109

:

or cursor.

110

:

I use cursor every day, and the argument

for using tier is the privacy piece.

111

:

The security piece,

and that it's an AI native IDE

112

:

versus vs code, which was an IDE.

113

:

And then they kind of chucked AI into it.

114

:

Am I getting that kind of right.

115

:

Yeah, it's it's about privacy.

116

:

And so it's so transparency I would say

transparency and seeing a customizability.

117

:

I think when you walk in this domain

of embedded software, IoT and so on,

118

:

you need to customize your environment

to your specific needs domains.

119

:

You need to connect with hardware.

120

:

And you have

you need to mix web based or cloud

121

:

based systems with local systems

because you you have the test environment.

122

:

Most of the time you need to run it

locally.

123

:

All these use cases

need customization in your toolchain.

124

:

In general.

125

:

You say you see,

I say toolchain and not IDE.

126

:

It's not only an industry

or my perspective.

127

:

Most of the successful tool

that we see today, first of all, as you

128

:

said, are ever proprietary

or single vendor tools.

129

:

But beyond that, it's very,

very difficult to customize.

130

:

Then if you want to make these kind of

131

:

I read deployments

that are local, NLM or local tooling,

132

:

plus a web based and cloud based tooling

and makes all of them in your toolchain.

133

:

It's very tricky,

and basically most of them,

134

:

you cannot do it

because you fully depend wander.

135

:

And this is where open source makes sense.

136

:

Here.

137

:

You can as soon as you will need

to customize

138

:

on your environment,

on your domain, on your test.

139

:

Yeah. Okay.

140

:

You see, what

this is for me is a key differentiator.

141

:

And why and so enthusiastic

about even bicep.

142

:

Yeah yeah.

143

:

And it's tracking for me now.

144

:

I mean I watch the YouTube piece

that you had the TM,

145

:

but this idea that we're kind of looking

at the classic open source piece where

146

:

and on the one side you get open source.

147

:

You can do anything you want with it

transparent,

148

:

you can customize it as much as you want,

go crazy because it's all right there.

149

:

And usually that's more difficult to use,

150

:

but more powerful is typically that's

my experience is like a neo non nerd.

151

:

And on the other side

you've got stuff like cursor

152

:

or someone like me who doesn't have

the coding experience comes in

153

:

and I can say like, all right,

I'm going to do whatever with IoT,

154

:

whether it's figuring out a codex or what

are codex with the LoRaWAN or whatever.

155

:

Is this kind of the big general piece?

156

:

Yeah, yeah, exactly.

I see you. Yeah. Okay.

157

:

You you give the example of course of

they are very powerful tool and for sure.

158

:

And that's why we see such a huge

and increasing traction

159

:

to all these tool first,

but also to on TR and open V6.

160

:

I was looking at the metrics.

161

:

We have something like a ten times

the factor grow on open

162

:

V6 traffic today we we this month

we expect to have more than 1 billion

163

:

of download extension download

compared to 100 million last month.

164

:

So you say in just one month the usage.

165

:

And this is because all this tool,

like you saw that you mentioned,

166

:

but not only cure.

167

:

So I can mention many others heavily

depending

168

:

on these extensibility mechanism.

169

:

And they're relying on the translation

open v6 marketplace.

170

:

Yeah okay.

171

:

So with that is the background of what

what it is and kind of how it works.

172

:

I keep an eye on the trends in IoT.

173

:

And the thing I'm seeing now is that

174

:

a lot of people are asking this idea,

like, how is IoT data being used by AI?

175

:

And that starts to be something

that is well within your wheelhouse,

176

:

where you say like, okay, here's

177

:

how we see people

potentially using their in their IoT

178

:

kind of workflows, whether it's

collaborating in real time or your ability

179

:

to run locally, any of those things

kind of spark ideas for you to say,

180

:

hey, Theia is a good use case

or good, good tool if you have X, Y, or Z.

181

:

Yeah,

because as I said, it's not only an idea.

182

:

You of course you say,

okay, it's known as IDs

183

:

because IDs, one of the application

184

:

that has been developed

with what we call the Theia platform,

185

:

and we are exactly the same model

as the eclipse ID was in the past.

186

:

Eclipse ID was not only an ID,

it was a platform.

187

:

And we have exactly the same model 40.

188

:

Yeah. Okay.

189

:

So what is the perfect customer

190

:

or really user of Theia

look like for you in the IoT space?

191

:

Like what?

192

:

What will they need

that only Theia can give, I think.

193

:

Yeah, really?

194

:

The part where I really believe

it's that we have a need.

195

:

TR is about setting up development

and testing toolchains.

196

:

Because as I said today,

if you want to, to take into account,

197

:

first of all, your environment

constraints, plus privacy and the question

198

:

on your dependency

some of the time, from what I see

199

:

from the industry, people don't

really care today about vendors looking

200

:

actually,

201

:

because they know that they are

they are maybe taking some risk or

202

:

relying on a single vendor, open source,

or even an appropriate solution.

203

:

But at the end of the day,

tools are so powerful.

204

:

That's okay.

205

:

They are ready to take this risk.

206

:

And they say, okay, let's go for it.

207

:

But specifically on in the field of IoT.

208

:

And again, I am not an expert,

but from my past experience,

209

:

or even if with the current users

210

:

or people who are members of the

of the working group simulation,

211

:

I see that

as soon as they want to to, to customize

212

:

and to make this hybrid configuration

with, with remote and local,

213

:

also take into account, as you say,

the privacy and security constraints.

214

:

Then they are

215

:

basically forced

to look at something like TR.

216

:

So this isn't the only solution of course,

but at least with TR okay,

217

:

we can really look at how it is

done and check

218

:

and make sure that they will put in place

this setup.

219

:

And that's why I for me

the key word is toolchain.

220

:

It's you have a tool chain.

221

:

You you need to not only build

and develop your components,

222

:

but you need to integrate with hardware

and software into your your environments.

223

:

And for sure, if you have to do it today

224

:

and from what I hear

from the people working in industry,

225

:

not me, but from from them,

there is no perfect solution.

226

:

Consequence

is that today the way they work is

227

:

they are using the powerful tool,

even if it takes the risk to be locked,

228

:

but it does not cover the full toolchain

and the full use cases

229

:

that they need to take into account

to put in place and from the development,

230

:

testing and integration environments.

231

:

So for me, if you want to take

the global picture and global touch

232

:

and problematic, it's good to have a look

at at this kind of project.

233

:

Yeah.

234

:

I mean, the more I hear about it,

the more exciting it gets.

235

:

Or it's like, okay, this is I can see it

being this super powerful thing.

236

:

If it's more than just this IDE,

if it's this whole kind of basically,

237

:

it's the AI tree

238

:

that you're going to hang all your stuff

on, whether that's interacting with it

239

:

and writing code or having the thing

do whatever it is that you want.

240

:

Let's talk a little bit more

241

:

about the customization piece,

how to handle prompt customization.

242

:

That's a really big thing,

certainly for for me as I go through.

243

:

Is that something that you guys

are focusing on paying attention to?

244

:

Of course, I don't say

it's it's a it's an easy topic,

245

:

but for nobody, it's it is actually

if you you can of course you can generate

246

:

you can do some via coding

and and get some great improvement

247

:

in your productivity, new software

development, productivity.

248

:

But at the end of the day,

if you are building a business,

249

:

you have to take into account

these requirements.

250

:

And so then you have the choice

between okay,

251

:

developing a lot of things by yourself.

252

:

And again

I don't say it's an easy task for nobody.

253

:

It will be.

254

:

Yeah.

255

:

But integrating end to end software

on that way or a system and testing system

256

:

design testing and and even integration

deployments

257

:

is the kind of thing

you have to do anyway.

258

:

So if you have just basically

you have the choice of having

259

:

just a very small part of your issue

solved and giving you an improvement

260

:

in terms of productivity

on one side or building.

261

:

Yeah, your your business

with full and on integrated environment.

262

:

But yeah, you are free.

263

:

Yeah I understand what you say

when you say okay, it's for me.

264

:

It's, it will cost maybe me a lot.

265

:

If I want to start.

266

:

You will.

267

:

My next startup.

268

:

Maybe you will pay

an additional cost at first.

269

:

But in that case, you can use to add

that to it.

270

:

It's basically is ready to use it.

271

:

And really what you can really compare it

with overrides

272

:

like so windsurfer, etc..

273

:

Sure.

274

:

So can you talk me

through the collaboration tools, the

275

:

the ways that people have to work together

with things

276

:

that might be different

from something they've seen before?

277

:

Yeah.

278

:

The collaboration tools, you mean? Yeah.

279

:

We have a project which is integrated with

to called the Open Collaboration Tools.

280

:

Open collaboration tool is really a way

for multiple developers to share

281

:

the screen and develop and make, code

review, real time collaborative edition.

282

:

So yeah,

this is a kind of additional features.

283

:

And this is a separate open

source project, by the way.

284

:

But like most of the AI features, it's,

285

:

combination of various open

source project and you can integrate them

286

:

all together

to, to bring these kind of new features.

287

:

Yeah.

288

:

And beyond that,

I don't know if you exactly what you,

289

:

you expected some on this side,

but for me, what is interesting

290

:

behind this kind of open source project

is for sure you have, this foundation,

291

:

this opportunity to, to, to to enable this

multiple, companies collaboration.

292

:

And it brings a lot of value.

293

:

But because there we have

they are all bringing their requirements.

294

:

And of course, sometimes you

you will find somebody ready

295

:

to implement your, requirement.

296

:

But most of the time the answer from

the community will be, okay,

297

:

you want this. So feel free to to start

298

:

the implementation

and feel free to contribute to it.

299

:

I mean, it is super exciting

to think we're living through this time.

300

:

That's this massive explosion of AI,

and there's all these tools out there,

301

:

and this is one of them.

302

:

And it's really important for businesses

to look at their practices and say, hey,

303

:

how do we want to integrate AI into

what we're doing?

304

:

And certainly for IoT businesses, it's

a big part of it.

305

:

I know I've talked to

306

:

whether it's the kind of one man shows

or the really big organizations.

307

:

They're all saying, hey,

we're we're integrating the saying and,

308

:

and there's a huge push in this industry

309

:

to make sure it's fully integrated

because it's such an advantage to use it.

310

:

Well, move your last use case.

311

:

And I'm thinking of in terms

of thinking of this industry,

312

:

is what I see more

and more people who say, okay, of course

313

:

we need lightweight models

and not only local model is not only

314

:

a question of privacy, but a question of

of size and resource consumption.

315

:

Yeah.

316

:

And I see some people from the community,

from today's

317

:

community

who are experimenting this to see, okay,

318

:

oh, can I manage to have local models

and then to bring these AI features

319

:

this time, not only for my development

environments, but for for my products

320

:

and to integrate into my product

some new AI features.

321

:

And typically in that case,

you are typically in the kind of,

322

:

use case

where you need to be able to modify

323

:

and to to tune the source code and,

and the prompting.

324

:

The it is, fully it's fully something

325

:

that should be done collaboratively

326

:

because, yeah,

if you rely on a vendor all time,

327

:

the prompting part is the circuit party,

the part which is really hidden

328

:

beyond the get the see, it's working

well again, it's great tool.

329

:

But yeah, if

330

:

you need to customize your prompting,

especially if you want to target

331

:

this kind of specific deployment

with lightweight models and local models.

332

:

Yeah, it is rad to see.

333

:

And there's all these little constraints

that the IoT side has, especially LoRaWAN,

334

:

where

there are the low bandwidth constraints

335

:

and your resource constrained

and the rest of it and saying, okay, okay,

336

:

we're not just working in that in an IDE.

337

:

There's this entire tool chain,

as you're saying, to say, hey,

338

:

from from soup to nuts,

how do we integrate this whole thing?

339

:

That's got to be pretty exciting

to work in there.

340

:

So very cool. Thomas,

thanks so much for making the time.

341

:

I know you're really busy with clips.

342

:

Thanks for coming on and explaining

a little bit about what the is doing.

343

:

Thank you very much. Have a good day.

344

:

That's it for

345

:

this episode of The Business of LoRaWAN.

346

:

I built this for you,

the one person in about 100,000

347

:

who actually has an interest in how

this tiny little slice of the world works.

348

:

Of course,

this isn't just about you and me.

349

:

It's about everyone in law.

350

:

When and how we can work together

to make an exceptional thing.

351

:

LoRaWAN is a dispersed community

with little pockets of knowledge

352

:

all around the world,

and most of them don't

353

:

talk to each other as much as I'd like.

354

:

So the first and best thing

we can do to make this show better

355

:

is to get more guests

on who I don't even know exist.

356

:

I want to talk to strangers.

357

:

Strangers who are your friends.

358

:

Please

introduce me to the most rad LoRaWAN air

359

:

you know, or point on my way

or help reach out and give me a name.

360

:

When it comes to running down

LoRaWAN guests, I can track a falcon

361

:

on a cloudy day

if you can remember metsci.show,

362

:

you can find me that's M-E-T-S-C-I

363

:

dot S-H-O-W, metsci.show

364

:

Okay, so sharing knowledge

by getting great guests on is the first

365

:

and by far the most important thing

we can do to make this better.

366

:

The next best thing for the show to do

is the usual stuff.

367

:

Subscribe to the show, give it a review.

368

:

Share it in your corner of the world.

369

:

Again, that's Matt's sideshow.

370

:

Finally, if you want to support the show

financially, you can do that over

371

:

at support.metsci.show You'll see options

there for one time donations.

372

:

If you really like this show

as well as an ongoing subscription option.

373

:

If you think the show is worth supporting

for the long term.

374

:

If you want to try LoRaWAN for yourself,

sign up for a MeteoScientific account

375

:

at console.meteoscientific.com and

get your first 400 data credits for free.

376

:

That's enough to run a sensor for

about a year if you're firing every hour.

377

:

The show is supported by a grant

from the Helium Foundation

378

:

and produced by Gristle King, Inc..

379

:

I'm Nik Hawks.

I'll see you on the next show.

All Episodes Previous Episode

About the Podcast

Show artwork for The Business of LoRaWAN
The Business of LoRaWAN
Learn From the Pros

Listen for free

About your host

Profile picture for Nik Hawks

Nik Hawks

Incurably curious, to stormy nights and the wine-dark sea!