


ModelFront (CEO Adam Bittlingmayer) explains why translation is “solved” only with the right context, and argues enterprises still need AI verification and workflow integration. The episode announces ModelFront’s free, AI-native OpenTMS, designed to eliminate “shadow localization” caused by legacy TMS friction.
Adam Bittlingmayer is CEO and co-founder of Language Tech Platform ModelFront; previously maintained Google Translate/Chrome integration. Based in San Francisco.
LLMs can translate many languages, but enterprises can’t reliably provide enough context; verification is the remaining bottleneck. ModelFront is “AI to check and fix AI translations,” taking responsibility for quality and engineering. Outcome-based pricing pays only for translations it approves (auto-confirmed), not for rejected segments. OpenTMS launches under BSL 1.1 (open-source path) with no lock-in, supports any AI provider, and is MCP-enabled for chat/UI parity.
Enterprise localization workflows add an AI post-MT check before human post-editing; ModelFront auto-approves more via automatic post-editing. Example customer: a medical devices design partner managing 26,000 curated terminology terms.
Written by AI. May contain mistakes.
Adam provides updates on AI translation and the challenges in enterprise settings.
“We were starting these enterprise pilots, rolling it out to those first customers.”
Adam explains Model Front's approach to improving translation outcomes.
“So I think we kind of did see that generations are getting better and better.”
Discussion on how Model Front interacts with localization teams in enterprises.
“And so they don't necessarily, they're not necessarily able to take advantage of the benefits of these kind of newer style models versus a specific MT engine.”
Exploring unexpected developments in AI translation over the past few years.
“So like, you know, I'm an AI engineer, super interesting to me.”
Discussing the scalability and chaos in enterprise content pipelines.
“And that's what a TMS is for, essentially.”
Exploring the concept of charging based on approved translations instead of word count.
“And it's like if something saved millions of words, but it was just a little bit worse than we thought it was going to be 12 months ago, that's still a success.”
Detailing the features and philosophy behind the new AI-native OpenTMS.
“It's just super, super easy, which it should be.”
Explaining the significance of Model-Chat Protocol (MCP) for ease of use in TMS.
“Even stuff like settings and stuff you can pull out.”
Learn about the upcoming launch of an AI-native TMS and its objectives.
“So sign up for the waitlist or request entry to the early access program at openloak.com.”

Get the gist, explore an idea or ask about something said.
On SlatorPod, what are the key points of this episode?
Automatic transcript. May contain errors.
0:00We built an AI native OpenTMS and we're giving it away for free.
0:10Hey everyone, and welcome to another episode of SlatorPod. Today on the podcast, we welcome Adam Bittlingmayer. April 2023 was the last time we had Adam on, so it's been over three years. It's time to do this again. As you may know, Adam is the CEO and co-founder of Language Tech Platform Model Front. Hello, Adam, and thanks so much for joining today. Good to be here again. Where are you recording this from today? San Francisco, California. There you go. That's where you want to be. We were there three weeks ago, four weeks ago. So yeah, now you're there. Cool. All right. So give us an update on what's happened since our last pod in 2023.
0:44If you kind of time traveled back then, sort of QE was just sort of coming out of research then. We were starting these enterprise pilots, rolling it out to those first customers. I guess you could say what happened was sort of the industry also sort of cut on QA became a thing. I think it's fair to say that a lot of folks had a lot of bad experiences with some of the QA features out there, right? Sort of random number generator, et cetera. We really built this track record of making it successful for enterprises by taking responsibility for the outcomes. So that was sort of a key to what we've done.
1:24And then we then wanted to improve those outcomes. We launched automatic post-editing, right? So we can auto-approve more. We shifted to the outcome-based pricing, right? So that was all sort of to improve the outcome. And then again, what we're going to talk about today now, this open AI data of TMS is what we've been working hard on lately. Yeah, unbelievable. You're mentioning the word TMS and I do want to talk about it, but first I want to start with the hard question. because you've been in machine translation, AI translation for quite a bit of time. So I want to start with the hard question, like where do these kind of like AGI level AI systems today still fall short, if at all, in terms of translation?
2:06You know, we had the conversation in San Francisco a few weeks ago was partially also like translation is basically solved. So where do you stand in kind of these, you know, godlike systems now and how they do it in terms of translation? I would say yes and no, right? It is solved in the sense that if I give it enough context in the right prompt, then it will spit out the right translation. It turns out that giving it enough context in the right prompt is actually really, really hard. Right now we're working with enterprise. I'm specifically thinking about these enterprise localization scenarios and just the number of dimensions that you have to deal with in an enterprise, like product, language, et cetera, is quite high.
2:47So basically, yeah, that turns out to be the problem at this point and the way to solve a solution. Should we still call it machine translation? We kind of started just calling it AI translation at Solitix. Machine translation starts. Yeah, I think it kind of always was AI translation. Machine is maybe a little bit more researchy, right? Machine intelligence, thinking machines and all of that. But if the person you're talking to understands AI, better say AI. Do you think we're going to call it super intelligent soon or is that going to fall flat? I think in some ways it already is super intelligence in the sense that nobody can do a hundred, no person can translate a hundred languages, but many of these LLMs can.
3:29Right. So. Okay. Look, I mean, you did kind of tease it a bit in the beginning, but I do want you to give our listeners and viewers kind of the elevator pitch for Model Front again. You mentioned, you know, giving it enough context, you know, making it work for the enterprise, et cetera, but yeah, give us the one minute kind of short pitch for Model Front. So Model Front is AI to check and fix AI translations. And that's used to basically skip manual translation for millions of segments that would not get edited anyway. AI, basically AI to check AI translations, right? And we're specifically focused on high stakes content, right?
4:04So actual like human localization workflows, where the outcome matters, and then we take responsibility for the quality, and for all of the engineering, right? So there's no engineering effort. You don't have to calibrate thresholds. You don't have to worry that the quality is going to drop. So you're going quite deep into like these enterprise problems, right? You mentioned giving it enough context and this being one of the key problems. Kind of walk us through kind of from maybe lead to implementation, like how you go about this? Who do you work with typically at a big enterprise? So we're basically working with the localization team, right?
4:38I typically head of localization and a lot of these orgs have like a right-hand man or woman who's responsible for AI, right? Was for MT and now for AI. And we end up often working with a lot of people across the team. It just depends on their style, right? If they're more of a Teams org or a Slack org. We go in there, we understand all of their workflows, content types, et cetera, et cetera, and then just make it work. So we basically help them get a step added in their workflow where AI, after the MT step, before the human post-editing step, AI is checking and maybe approving some of those translations.
5:21If it rejects, it might edit some of them and then that gets checked again. Maybe more of those get approved and those are basically auto-confirmed inside of the workflow. Is there a specific kind of vertical or language setting that is more challenging than other settings? Or is it kind of broadly, it's just an engineering challenge combined with a language challenge, combined with an AI challenge, and it doesn't really matter which vertical you talk to? It's pretty vertical agnostic. I mean, you can definitely get more acceleration in some verticals, right, if you have easier content. But I don't think it's so much about the content type itself.
5:58It's not like what's in the text is harder. It's like the fragmentation itself, right? What we were just talking about. So if you have something that's just this like very narrow, like, you know, very restricted domain with, you know, only the same few hundred words that are reused again and again, like you don't even need AI to kind of automate that, right? If you have something that's where it's super one-off, every single thing is this, you know, like a new game or whatever, that's pretty hard to do. No, most people are somewhere in the middle. So you guys call your company Model Front. years ago, still a great name for a company in the AI space.
6:34If you named it now, you probably, you'd wanted to call it Modelfront, Modelfront.com. So is your original thesis kind of still working out as planned? Better AI means more demand for Modelfront? Yeah. So I think we kind of did see that generations are getting better and better. There's going to be a need to get verification and you also have just more and more content that gets generated. it's generally, that has been very, a tailwind, if you want to put it like that. Is there anything that like majorly surprised you in how things went over the past three to five years? Like you probably, I mean, you had a deep understanding of these models, right?
7:12And okay, there's more scale, there's more compute and it kind of scales up and it works much better, but is there anything that kind of, yeah, surprised you in those past four to five years of the boom? I do not want to say that I predicted all of this so to speak, right? I think when you come to it again and again, is that if you could play with the models we have now, let's say two years ago or five years ago, if you could just play with one, sit in the room and chat with it or whatever, you would think, oh, it's done. It's solved. Like, you know, and I don't have to take out the trash anymore.
7:48I don't have to, you know, all of this stuff, it's just going to be done. And I don't have to deal with, you know, organizing my localization workflows anymore. All of this stuff is just like translation is going to be solved. And a lot of other things are going to be solved, too. And then somehow we're here and it's not right. And, you know, as you hinted, right, I do think and have always thought that, OK, generation gets really good. Now we need to verify when it's actually correct. That's the thing that lets you kind of rely on it. That said, I still think that's kind of what would always surprise you.
8:23I think anybody would be lying if they said, I would go back two years or five years and work with one of these things. I would not predict well what the remaining bottlenecks would be, so to speak. That makes a lot of sense. Yeah. So Model Front doesn't need to create the original or generate the original translation, right? We don't, right. Which models are you working with the most? How are the specialists doing versus the hyperscaler versus the Frontier Lab LLM output? I mean, they're all kind of using the same stuff under the hood, but I think it's very clear that with a chat kind of model, you could pass it way more additional context, right?
9:00You could give it tons of conversations and documents and so on and basically give it a more ideal prompt and context and get a better translation. If you actually look in the kind of localization workflows that we work in, our customers, because of just the complexity of the enterprise that we mentioned, plus sort of the way that the existing software works, it's actually not really possible or certainly not easy to pass all of that context to the LLM. And so they don't necessarily, they're not necessarily able to take advantage of the benefits of these kind of newer style models versus a specific MT engine.
9:39And so we do see that although there's a slow shift, let's say, towards the LLM style, it's a slow shift because it is just easier to use something that just works, namely, you know, one of the MT engines. And the MT engines are evolving as well. So it's not as crazy of a shift as you might think. So there's no massive pressure from like the kind of senior level, the C level to like, okay, why are we still using specialized things? Isn't this all kind of solved now? Like, okay, we need a layer, but why do we need two additional layers? Why do we need a specialized model and then something like ModelFront on top of it?
10:16The MT providers like very clearly do not provide something like ModelFront. So that hasn't been a big issue. And I think, again, at that very high level, they don't necessarily distinguish between like a MT provider and then the AI for generating the raw translations. So you guys look at, like you adapt models, specific customers, language pairs, domains, content types, terminology, quality, et cetera. So how does this work? Like who does what work here? Like you do all of it, you come in and like, you know, you talk to the head of localization and you basically take stock of what's there or yeah, how does that work?
10:51We kind of touched on it a little bit earlier, but maybe you give us some more color to that. So yeah, we take full responsibility for that. I guess that's a bit unique, right? We talk to them, but we really look a lot at the data because, you know, sort of you step back, let's say the style guide says one thing, what people actually do often says another, and it's not always wrong. Oftentimes it's the style guide that's out of date, right? So you've really got to go look in there and you can find out all sorts of stuff that people inside the org don't know about what's happening. So we do all of that.
11:24And we're basically taking all of that complexity that exists, naturally exists, and you can't get out of it. They're not doing anything wrong. It's just you have this many departments and this many content types and this many quality tiers and this many languages and the thing just explodes. right? Taking that and distilling it so that they just get the thing at the end, which is, okay, I saved, you know, 50 million words, 50 million words, we automated with AI, 50 million still have to go to manual translation agency, for example, right? So just distill all that. The way I think about it is like Waymo, right?
11:55So like, you know, I'm an AI engineer, super interesting to me. Still, when I order a Waymo self-driving car from the app, I just want to get from point A to point B. And all that complexity is under the hood and they have live operations and all this stuff. I don't want to know about it. I just want to get from point A to point B. And so that is what ModelFund does for enterprise translation. Yeah, Waymo. Took it for the first time in San Francisco four weeks ago. Blew my mind. It was interesting. After two minutes, I got used to it. I'm like, that's it. That's the future. You get really used to it.
12:25Yeah. Let's not go back. Like five minutes. I'm like, I'm trusting. It's safer. Yeah. It felt very safe, right? Right. All right, one more kind of like recent phenomenon that I've seen is these kind of embedding of AI translation in ChatGPT clawed by, I guess it's called MCP, right? Model Context Protocol. So basically you're in your chat interface, you connect via your kind of AI chat to these specific language tech platforms. So are these kind of language tech platforms in danger of becoming like super invisible behind that interface? And what's the value at there still? Or is it more of a playful thing right now?
13:06You kind of want to be part of the conversation. You want to be where people are chatting with these frontier models. So I think there's definitely still a place for TMS. I have a lot of experience working on the other side. So I was actually maintaining that Google Chrome, Google Translate integration, which is, right, if you kind of go back, of MT got embedded in browsers, MT got embedded in social media, et cetera, et cetera. Well, those things were never going through a TMS, right? And they're incredibly useful and incredibly impactful for people who just need a translation right now. They don't have to copy, go to Google Translate, paste, and come back, right?
13:40Or whatever. So that's always been happening. It was always, I'm going to conservatively say, a million times more volume than the sort of localization industry, but it's probably even in the billions, right? And so that's just getting easier and easier, but it's not like a zero-sum thing. That just makes it much easier for people to get that in the place that they're working, and that's unstoppable, and it's a good thing, etc. But I would say for these sort of high value, like enterprise content pipelines, anything with a chat, if it's in a chat and someone's interacting in the chat there, that's like already too slow and unscalable and chaotic for the kind of content that we're talking about.
14:23And that's what a TMS is for, essentially. I mean, we can only speculate, but basically these companies doing these, the kind of LTPs, as we call them, they're doing this just to remain relevant, to be part of the conversation. You don't want to not have this, but it's not like, yeah, essential if you're in this super high volume, highly automated content pipeline ecosystem. Right. It creates another way to get it, which is good. It grows the pie, so to speak, from my view. You mentioned outcome-based pricing before. So you introduced that apparently in, yeah, this year, I guess, right? So for translation that it successfully automates rather than every word process, so you charge for translation that is successfully automated rather than for every word process.
15:04Yeah, tell us more. So if you think about how MT and now AI and human translation is, it's just sort of by the token, right? It's by the words processed. But you look at what Monofront does. We approve some translations and we don't approve other translations. So the most aligned thing with our customers is they pay if a translation is approved, they pay us, but then they don't have to pay for manual translation because that translation got confirmed. If we say no, we can't sign off on this, they pay us nothing because they still have to send that one to human translation. And so that way you have this thing where, okay, I don't have to know exactly.
15:41We don't know what our volume is going to be this year. We don't know exactly how much we're going to save. We don't know if our NT engines are going to improve or our content will get more difficult or whatever this year. So I'm making this purchasing decision. And it doesn't matter if we save 50 million words or we save 40 million words, we're going to come out with a positive ROI at the end of the year with this structure. And it's really rare for AI. But you do see it in some other verticals, but certainly in translation and QEA, it's very, very rare, but it's very aligned with customers. So they get it?
16:16Like new customers? They get it. We started that two years ago. These are long enterprise contracts, so it took a while. But we shifted everyone over and much better for them, which is also better for us. So was this something that came from kind of customer conversation slash expectations or you thought this was going to be better and you launched it, they liked it? We had been talking about thinking about this for a long time. And then, you know, sometimes things work out better than expected. Sometimes they work out a little bit worse. And it's like if something saved millions of words, but it was just a little bit worse than we thought it was going to be 12 months ago, that's still a success.
16:56and why can't that be right? And so we had always kind of wanted to do it and that was the push to do it. And everyone we talked to said, actually some did resist a bit, but because it was like a little bit new, but they're all happy that they did it. All right. So, you know, we are here to also bring breaking news. So you are launching an open TMS and you make me say the words TMS because we, you know, basically rechristened all of these companies LTPs. But actually, you're kind of making the case for the LTP because if an MTQE company now also launches an OpenTMS, what are you going to call that company?
17:33So it's an LTP. Anyway, that's my little opening here. So an OpenTMS, why, how? We built an AI native OpenTMS and we're giving it away for free. To be clear, so it's not a business for us, right? We'll provide hosting at cost, but it is not meant to be a business. Like that is just fundamentally changing now. Yeah. And we basically do believe in the idea of a TMS, right? Like the TMS is a good thing. We can get into why, but the way the status quo, the way that some of that software has been operating is creating a lot of frustration. And so we believe that open and AI native will fix those problems.
18:21So what is the problem with the status quo in TMS in your view? So you touched on it, right? I mean, what is happening inside of these enterprises is you have a content team, like say the marketing team, they have their content that they want to translate. And they would like to send it to the localization team. They'd like it to have the terminology applied, somebody else responsible for, you know, dealing with the LLMs, they'd really like that. but there's also some button in their CMS where they can just get an AI translation. And it might not be great, but it's just one button, right? Or checkbox.
18:56If you create too much friction, they will do that. And it doesn't come through the TMS. It doesn't come to the localization team, right? It doesn't come to the AI that the localization team manages. And that's what's happening. And the reason is that it's understandable. It's really understandable. software built 20 or 30 years ago was not built for AI. In fact, AI is even taxed. There's an AI tax, right? And you're not incentivized to put like mass amounts of raw MP through this, which is a need inside an enterprise, right? And a ton of little things like that with regard to integrations and so on and so forth.
19:36So this is all existential, I'd say. It's problematic, but it's existential. So, yeah, these are the problems that we see most teams have right now every day. And we've had these conversations at SlatorCon San Francisco three weeks, so exactly that, right? So, you have this, what we call, I mean, either translation is a feature or language AI is a feature that kind of shadow localization, the translation button, whatever you were going to call it, right? It's a big temptation to use that. And what you're saying is if the TMS component is just not native enough, not easy to use enough, it's not going to get used, right?
20:18So it's a bit of a race to make it easy. And you're saying, well, so we can just as well do this on our own. We can offer this. We can give it to clients and they can choose to use it or not. Exactly, right. Because there is a ton of value in the TMS, right? I know it's really become fashionable to say no segments, no TM, no per word pricing, all of this stuff. But if you think about it, what does a TMS do? You reuse content that's already human reviewed. That makes a ton of sense. I mean, if I have a billion words, but like, you know, 90 % of it, we've already human reviewed. I don't want to send that all to an LLM right now.
20:57That makes no sense, right? We are managing terminology. The design partner we work with this on, they have 26 ,000 terms that they've curated over years, right? And it's medical devices, right? So it really matters. So they manage all of this. They're the people who know about this. And this goes to the AI, right? So every single thing you can think about, it makes sense for it to be managed centrally inside an enterprise, right? And there has to be a place to do that. There has to be really powerful tools for the team who manages to do that. So you mentioned open and AI native. Give us more specifics.
21:34What does open mean? What does AI native mean? They're actually pretty hard to disentangle from each other. I think open, in a lot of ways, leads to AI native. So firstly, just no lock-in, right? It's launching with a BSL 1.1 license, which converts to open source, procurement loves it, etc., etc. No limits. And, you know, you just can't put limits because anyone can just, if they really want, can go and host it themselves, right? And AI native means, of course, none of those taxes. Just use any AI that you want. You can use it at high scale, et cetera. Also stuff like prompt building. So it's really easy to build prompts.
22:15You can see the prompts that are being sent to AI. You can edit them. You can even do QE, let's say, for an XML validator or spell checker or stuff like that. It's just super, super easy, which it should be. because these models, we all know how to use them, et cetera, et cetera, but it has to be in the place that you do your work. I think the other thing that's really game-changing is full MCP. So meaning that like anything I could possibly do from the UI, I can also do from chat and I don't have to do it from chat inside. I could never open this TMS. I could do it from my enterprise approved Azure chat GPT or Claude or whatever I choose.
23:00uh securely and i can do stuff like okay in this workflow like advance all of the files to the next step or look for any any any time we use this word that we shouldn't use or build me this dashboard and then that dashboard is there and and it's deterministic right so just whereas if you think about the old world you had to request a dashboard and wait for a dashboard and pay for this dashboard and all of this stuff is just going away. Did you feel now is the right time because we're kind of hitting like, I mean, I felt like it's a bit of a step change in the past six months. So like you could have tried to do this maybe two years ago, but do you feel now is a better time?
23:43Like, yeah, like MCP wasn't a thing, I guess, at least not in my head two years ago. So going out now is you feel like this is going to get, it's going to have legs in the next few years despite all the progress that's happening. Yeah, I think if anything, if we look back, I think you and I have been too early on a lot of stuff. So yeah, I do think it's the right time. I mean, I just think that people have this problem. They need to solve the shadow localization problem. They need to use AI inside of localization, right? And so the problem is now, and I don't know, I'm very skeptical of the idea of like, do nothing because this or that, you know, the world's going to change in five years, so to speak.
24:33If you always do that, if you always think like that, you'd never do anything, which is clearly not the right answer looking back. Depending on your life philosophy, man, like maybe sometimes you just sit there under a tree. I mean, sometimes it does help to just wait. I hate to even say the word vibe code. It's not even vibe coding. It's just like prompting with natural language, which helps me a lot. I can do stuff that's really cool now that I would have had to actually think about six months ago. Now I can just kind of two prompts and it's there. Anyway, sidetrack. So it's usually very hard to migrate from like an embedded TMS.
25:09Like how do you facilitate this if people are interested? I sound like a broken record here, but MCP. So this is really changing how we operate, right? So basically every platform has MCP or has to have MCP now to be table stakes. And so just the whole pain of migration is becoming radically easier. Even stuff like settings and stuff you can pull out. So we're willing to help support folks with that. But basically it's becoming radically easier than it was, let's say, a year ago. So can you just explain MCP for those of us who don't have a deep background in AI? Because it's new-ish. I mean, again, I've heard it the first time maybe, yeah, I don't know, four or five months ago.
25:54Has this been around and now it's just kind of becoming like more mainstream? It is a little bit of a rebranding. So let's say APIs. We all understand APIs, right? Anything you should be, this is an old concept. And I'm going to sound really old now, but there was something called the data liberation front kicking around in Silicon Valley maybe 20 years ago. So basically anything that I can get out, that I can see in the UI, I should also just be able to get out easily via an API or MCP. MCP itself is just a protocol which lets AI, whichever AI you choose, easily get stuff in and out. And it's made for that app.
26:31So basically it takes care of stuff like roles, right? If you don't do this, what do people do? They just go have their AI go to the browser and actually use the browser and get the data out. That's really hacky and wrong. And so the right way is to just accept the fact that people want to use AI to do the boring mechanical tasks and open it up for them in a safe way. And so that's the essence of it. Did somebody launch this? Where does this come from? It was an anthropic thing. so you know open ai sort of has their competing thing but has also embraced it so to speak but i i think mcp is maybe even if it were an mcp it'd be something else right like the point is that people are making it easy for agents to interact with software and so you don't necessarily have to monkey around with the ui anymore which means that some things that were a pain before are just radically easier and also means that if software doesn't do that it's just radically worse user experience than the software that does provide MCP or something like that.
27:36So if I'm interested, like as a localization manager, what do I do? How do I get started? So we are launching an early access program. And for those folks, qualified folks, we are providing AI credits, engineering resources, actually, to build a thing, free hosting, etc. There will also will be general availability coming soon. But you can find out about that at openloak.com. So sign up for the waitlist or request entry to the early access program at openloak.com. We'll put the link in the show notes. I guess this is going to be your priority for the rest of 2026 and probably early into 27, right?
28:18Rolling this out and, you know, both working with the folks in the early access program and then the general availability. But then, you know, on the other side, right, what is the goal in the end is making AI more accessible, getting AI to more folks and having better AI outcomes, basically creating more value with AI. So with this open TMS, with bigger models, with bigger context. All right. Well, sounds like a lot to do in the next few months. All the best with the launch and thanks so much for joining today, Adam. Thank you, Florian.
28:53Right now.
Adam Bittlingmayer, CEO and co-founder of ModelFront, joins SlatorPod to discuss the state of the art in enterprise AI translation and the language technology platform’s (LTP) launch of an AI-native open translation management system (TMS).
Bittlingmayer says translation can appear solved when an AI model is given enough context and the right prompt, but providing that context consistently within complex enterprise localization workflows remains difficult. Enterprise teams must account for different products, content types, terminology, quality requirements, languages, and existing systems.
The CEO revisits ModelFront’s core value proposition of using AI to check and fix AI, allowing enterprises to automatically approve translations that would otherwise require human review. Focused on high-stakes localization workflows, the LTP takes responsibility for translation quality and the engineering complexity behind it. He also highlights its outcome-based pricing model, under which customers pay only for translations that the LTP successfully approves.
Bittlingmayer also talks about why neural machine translation models remain widely used despite advances in LLMs. He says newer models can benefit from much richer context, but existing enterprise workflows and software often make providing that context difficult. He sees verification as increasingly important as AI-generated content and translation volumes continue to expand.
Bittlingmayer then introduces ModelFront’s AI-native open TMS, OpenLoc, which the company is making available for free, with hosting offered at cost. He argues that traditional TMSs were not designed for today’s AI workflows, creating friction that can encourage teams to bypass centralized localization processes altogether.
He says the new platform is designed around openness, flexible AI use, prompt building, and MCP connectivity, allowing users to interact with localization workflows through enterprise AI assistants as well as the traditional interface. Bittlingmayer sees this as part of a broader goal to make AI more accessible while improving enterprise AI outcomes.
ModelFront is launching OpenLoc through an early access program, with general availability planned to follow.







