Stop Rationing Tokens: Let the Harness Pick the Model — Kimchi by Cast AI

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Stop Rationing Tokens: Let the Harness Pick the Model — Kimchi by Cast AI · Channel: AI Engineer · 17:48 · Watch on YouTube

Topics: AI agents, Task and workflow automation, Software engineering with AI, Model inference and optimization, Local and open-source models

Summary

Cast AI co-founder Laura presents Kimchi, a coding agent designed to reduce token costs by selecting models according to task quality and cost rather than restricting developers’ usage. The team reports 2.5 times lower costs than cloud coding while token usage grew, and describes its open-source harness and long-running “ferment” process, which repeatedly builds and scores changes. They also demonstrate Teleport, which keeps coding sessions running in remote sandboxes, and Studio, a team interface for reviewing and sharing agent tasks. The workflow currently stops at staging to retain a human check before production.

Key points

The speakers argue that limiting developers’ token usage is the wrong response to rising AI costs; managers should make coding agents affordable to use extensively.
Kimchi’s harness selects a model for each task based on cost and the quality of the outcome.
After three months of use, the team reports 2.5 times savings over cloud costs while token usage increased.
The harness changed its preferred models over time, illustrating how automated selection can adapt to shifts in model performance and cost.
The open-source harness includes “ferment,” a process for long-running coding tasks that breaks work into milestones, checks results, and iterates.
Tasks are scored, and an artifact is considered complete once it earns at least a B; users can request further work toward an A.
The current workflow deploys completed changes to staging and keeps a human in the loop before production.
Kimchi Teleport runs coding sessions in secure remote sandboxes so they continue when a laptop is closed or disconnected.
Kimchi Studio lets teams share agent plans and work in progress, review tasks, and coordinate them in a Kanban-style interface.
The speakers say the harness is entirely open source, while Studio and Teleport require a Google account.

Chapters

Start	Chapter	Summary
00:13	Token costs and the case for model selection	Laura introduces Kimchi and argues that an automated harness can choose cost-effective models instead of limiting developers’ token use.
04:20	Kimchi’s cost savings and adapting model choices	Laura shares the team’s reported savings and shows how the harness shifted its preferred models over time.
07:24	The open-source harness and ferment workflow	Zilvinas explains how ferment handles long-running tasks, scores output, iterates on changes, and currently stops at staging for human review.
11:03	Teleport: coding sessions that keep running	The speakers describe remote sandboxes that keep agent sessions running in the background and make them accessible across devices.
14:09	Studio for team collaboration	Zilvinas presents Studio as a shared workspace where colleagues can review plans, coordinate agent tasks, and track work in progress.
17:18	Open-source availability and closing	Laura summarizes which Kimchi components are open source and invites attendees to see the tools.

Quotes

Our jobs as manager is to do the opposite. It’s to ensure that we want to make the tokens unlimited. — 01:45

Wouldn’t it be nice that you have an automated fashion, automated engine, automated harness that will select the right model for the task at the right time based on the outcome of the task? — 04:20

Artifact is only considered complete once like the output of the scoring is like at least a B as a score. — 10:00

It just doesn’t run inside your laptop. It feels like it. — 13:38

If they close the laptop, >> It still works. — 16:46

Concepts and entities

Kimchi, Cast AI, Laura, Zilvinas, Kimchi harness, ferment, Kimchi Teleport, Kimchi Studio

Full transcript (original language)

00:13 All right, guys. Very nice to meet you all. My name is Laura. I’m the co-founder and president of Cast AI, who is the inventor of Kimchi, which I’m going to show you in a few seconds. Um, the talk is going to be about preparatory models for edge case, but use open source for the rest. And you’re going to see some great we have with that with the coding agent called Kimchi. I have the pleasure to be here with Zilvinas. >> Hey guys, I’m Zilvinas. I’m leading engineering for a Kimchi coding platform. >> So, you have Zilvinas who run the entire

00:44 engineering team for Kimchi. I’m so excited to be here with you. All right, let’s go. So, um, you have you may have seen this in the news. Uh, two two piece of information that came up in the last few weeks. The first one is there’s a company in India that spent $500 million on Anthropic alone in one month. Which is again of a good example of tokens that are completely out of control. The other example is a famous tweet from Uber, the city of Uber, who say that in the last four months he

01:14 actually used the entire budget of Anthropic for the entire year. And this is how much the token, um, the token maxing or the token cost is a thing for the enterprises. And you’re going to see how we fix this. Companies responded this by limiting the use of token. You may have heard this many, many times with our friends developers. And as a developer, when you limit the amount of token, it feels like this.

01:45 It’s like, “Hey, you can use your laptop. You have an hour battery and you can only charge once once once a day.” That’s how it feels to limit the amount of token for developer. We at Kimchi think this is wrong. Our jobs as manager is to do the opposite. It’s to ensure that we want to make the tokens unlimited.

02:16 And inexpensive. Our job is not to prevent the developer to use coding agent. Our job is to make sure they can use it as much as they want for as long as they want in a completely unlimited fashion. This is why we started Kimchi. At Kimchi, we have 300 developers. Oh, sorry. 300 employees. 2/3 of those are developers. The Zivinas run that. And we’ve been using this coding agent for the last 3 months. I’m going to show you what we built, how

02:49 we built it, and the result of it. And then Zivinas will show you a few more things that make the coding agent really, really cool. It’s coming now. We built an entire token optimization based on this idea. What you see on screen is the true cost of LLM model for the same task. It was made by the famous white paper in a university. And

03:19 here’s what they say. On one side, you see the cost per token. On the other side, you see the cost per task for the same task with the same quality. That’s is the important, otherwise you cannot compare apple with apple. And here’s what is written here. First, they are all different. But also, look at Gemini 3 Plus. Our our friends at Google. Well, it seems to be a relatively inexpensive model at $3.5 per

03:50 million token. This is a blended average at the time this study was made. But actually the task to complete with Gemini 3 was $305. Sorry, $705. And look at the others. Minimax 2.7 is one of my favorite. The cost per token is 1.5 and the task the the cost for the task is $148. It means the models are not the same. They don’t cost the same, but

04:20 also not the same per task. And that’s why we built our coding agent. Wouldn’t it be nice that you have an automated fashion, automated engine, automated harness that will select the right model for the task at the right time based on the outcome of the task? That is what we did with Kimchi Coding. That is what Zilvinas did with his team. And here’s the result. We’ve been using

04:50 it for 3 months. 2.5 times is the amount of savings we have over cloud for the period. This is a graph of what would have been our cloud bill by day. The purple one on top is what would have been cloud cost over time. The blue line is what we paid. Same outcome. But look at this. The number of tokens

05:22 increased by 1.5 times. Okay, we all experienced that. The cost to cloud decreased by 1.5 times. This is our job as managers. We provide to our team a coding agent that is essentially unlimited because of that coding agent is growing 3x less, 2.5 times less than the gross of token for the task.

05:52 There’s another result. The multi-model I showed you before. I’m going to show you the evolution of the selection of the model over time by the harness. And here it looks like you see there is a lot of orange early June and there is a lot of yellow late June. There’s a complete shift, meaning the harness is selecting a very different model. It happened the 12th of June.

06:23 And this is what happened the 12th of June. The 3rd of June, Kimi Kimi 2.6 is winning by far the most popular of the harness. 21st of June, the one winning is Minimax 3. Also as human, probably we can find and understand and test and be comfortable with a new model coming in. Do you think we look at this as human? Absolutely not. But an autonomous harness, an automated

06:54 coding agent that is obsessed with token cost is going to do just that. That is what I mean when I say our responsibility is to think about providing our developers with a right cost per token for the same outcome. It is our responsibility to do it. Ziminus is going to tell you a little bit more of how it works because whatever you’re going to see,

07:24 it’s open source. Go, Ziminus. >> Okay, so one more thing to mention. So, while I mentioned that token maximum is no longer an option for everyone and uh as a manager at the company as the of you guys probably heard, the having unlimited amount of tokens to spend is no longer looking to be like a case and we had to build this harness for ourselves. So, meaning like we ended up spending too much on cloud code like half a year ago. We saw this

07:56 cloud bills skyrocketing and we started working on our own harness. Behind this is using like PyMon OS SDK, but the goal of that is like to take the open source experience, still contribute back to open source by providing our own harness to open source. Building constructs like ferment which fits into the kimchi theme which is meant for long running tasks where on average we estimate the human in the loop for these tasks is uh typically happening only like every 2 hours, every 3 hours and etc. So, the goal is uh

08:26 use kimchi harness, you run a ferment process, it asks you a bunch of questions and autonomously runs the coding task for like more than 2 hours. The goal of that is that because it breaks it into milestones, it knows how to implement it and at the end of the cycle you would get scoring for your output so you would actually know what’s happening. And going back to previously discussed like the amount of models used with uh kimchi harness, the scoring is there to prove you that whatever model we will pick for you, you will always have like

08:57 good enough scoring for the output on the task you’re implementing. The way whole kimchi platform works, the goal is like uh we are not just providing you the coding harness, we want to provide you the full software development life cycle solution. So, you do a change, it runs a build for you, it checks if something broke, if something broke, it will go back, do the change again, rebuild again and that’s how this fermentation process looks. If everything is complete, there is no extra remediation needs to be done, the changes will be deployed to the staging

09:29 environment. >> You see what what I like with the way you guys explain it is the harness select the model, ferment is constantly checking the quality of the code and going back, meaning there is a a higher order model that is checking the quality and thinking that quality is not good enough, please do it again. This is what I was saying earlier by it looks at the outcome. And it’s optimizing the token based on

10:00 the quality of the outcome. That is what is super cool with ferment. >> Yep. And artifact is only considered complete once like the output of the scoring is like at least a B as a score. And even for that, like if you don’t like the score, you can ask agent to go back, fix it up and like it will try to achieve a A as a scoring. The intention of stopping at the staging is currently by design. We still believe there’s like some little human that needs to be in the loop before going to the production at least at this point.

10:31 We all know that the ultimate future is to avoid that. You go to production, you look at the metrics, you look at the observability data, the SLO metrics you have in your environment. If you’re running on Kubernetes, you want to check if the pods are not crashing, if application health checks are passing. So, all of these things need to be done before it ship it ship to production automatically. So, this is like one of the things that are coming. So, the current limitation we stop at staging, but they’ll go to production and and further on that. What we believe is that reading code is

11:03 not enough anymore. So, like probably most of you know this like across this year like the more live engineering or agentic engineering we do in our coding environments, the more code changes we get in our change sets and it kind of becomes unbearable to read, especially as you keep bringing like a little bit of a less technical personnel into your pipelines. So, imagine like product manager doing the coding or my dear colleague Laura starting to do some coding and suddenly he comes up with a PR which is like 2,000 lines long and you need to

11:34 actually review not just the differences but the actual intent which he wanted to achieve and the specifications which he provided to the coding agent. So, yeah. Moving on. One thing we built more on top of that is the thing we called Kimchi Teleport. So, Kimchi Teleport allows you to close your laptop. It spins up sandbox in the remote environment. It’s a secure sandbox where you don’t need to care about anything. We can do that both on our SaaS platform. We can ship it to your on-premise deployment if you’re

12:05 like a financial technology company and things like that. So, we will manage these agent sessions for you. We will make sure they are running. They are accessible both from your laptop, from mobile devices. They are running in the background. So, it goes back and ties to the story that I mentioned about the human in the loop is only appearing like every 2 or 3 hours. So, Teleport emphasizes that. You run your process. You teleport it to the remote environment. The young spins up in the background. You don’t need to care about it. We will make sure it’s the right size, it’s the right security

12:36 requirements and everything else and that will keep on running until the task is done or until the human in the loop needs to be involved, get notifications for some clarification of specs and things like that. >> Guys, I really like this idea. Look at the this This is a screenshot of Teleport. I am puzzled that sometimes innovations are always the simplest one. Like this was born out of the idea that one of us was in the plane coding

13:06 and then Wi-Fi breaks. And what do you do when Wi-Fi breaks? It breaks the coding because it lives inside your thing. I think it was not the Wi-Fi. It was the battery that broke. Not the Wi-Fi. So, you lost So, he That’s it. The guy was stopped there. What Teleport does it it takes all your environment, it synchronizes with a container that live inside one of the hyperscaler. Now we can use Google. And locally on your laptop, it feels

13:38 exactly the same. It just doesn’t run inside your laptop. It feels like it. It runs somewhere else in your environment in a container in a Kubernetes cluster inside the place at Google that will continue running even if you close your laptop. How simple that is. And yet our favorite tool is this one. I think the the last one I I I looked at it, 62% of our engineers are only using

14:09 Teleport when they are coding now. For that problem, they go home and continues. They’re in transit, it continues. They’re on holiday, they don’t look at it anymore. It continues. That’s so cool, huh? >> Yep. So, there is more to that. So, we built another tool called Kimchi Studio. If that Teleport is something typically individual uses, Studio is kind of like a Teleport for teams and enterprises. The goal behind the Studio is that you have a camera like board on your

14:40 environment. It again runs in the Teleport sandboxes in the secure environment. The thing here is like you are supposed to share your tasks on work in progress and plans which are made by the Kimchi code harness with open source or proprietary models with your teams. So, for example, imagine you as an engineer you are like working on a task. You started implementing asking for a plan. The plan can be easily reviewed by your colleagues like by your peer engineer, by your product manager or someone else who cares what’s happening and this is how you kind of work with your

15:10 whole team on collaborating on this agentic engineering journey instead of like saying, “Oh, I’m going to type this prompt. I’m going to try to execute it to the production.” Only later on to learn that the specification and requirements were incorrect, uh the output is not what was actually wanted by the task and things like that. >> And this is yet another thing we built for ourself. We found it so cool, we decided to make it a product. How do you work with your colleague if what you have is a is a is a terminal, a

15:43 CLI? How do you share that with the others? Well, first he has to live somewhere else. So, Teleport is the base of what we call studio. And studio is a very nice way to visualize all the sessions that are running, including those that are asking for a review. And then someone in the teams can say, “I take it.” And I answer the I I answer the question that is for review. Then you answer it, it goes back to in progress

16:14 until the next one goes into review. And you can have many, many of those in a team in a pizza team, a team of five to 10 guys. Think about it. A yet yet a very simple innovation that is so essential in how our engineers are coding. They can now share in a very simple way, right? It’s a It’s a browser. It’s graphical. They can share and they can take over the task according to the Kanban style

16:46 of what is in review, what is in progress. They can start new ones with backlog. And you know what? If they close the laptop, >> It still works. >> It’s also cool. >> Yep. >> This is really cool. >> Thank you. >> And >> Yeah. We forgot to show the screenshot. >> This is the one that we took over, right? That’s the one. Guys, you can see this at our booth if you want to uh to see more of it. We are in UG4, it’s exactly this way across a few streets. But, have a look at it.

17:18 It’s really good. It is open source, most of it, at least. The harness is entirely open source. Studio and Teleport, you need a Google account. So, it’s a little bit more more complicated, but it’s a a 5-minute install. And then you get your Teleport set up, as many sessions as you want inside your Google account. And then you can share with share it with your colleagues with Kimchi Studio. So, come see us. We’ll be delighted to speak to you. Thank you, guys.

17:48 >> Thank you, guys.