adam_patarino

Fair enough! The simple answer is: we did a lot of work to make the model better at coding without requiring complicated installation or configuration. One comman to install and run.

All the benefits of claude code, without any of the limitations or rug pulls.

Its batteries included. No config.

We also fine tuned and did RL on our model, developed a custom context engine, trained an embedding model, and modified MLX to improve inference.

Everything is built to work with each other. So it’s more like an apple product than Linux. Less config but better optimized for the task.

Tell me more! Thanks for the waitlist

I’ve seen this story making the rounds and I’m not just why it’s gotten so much traction. Is it just a good write up?

You check the plan files into git? Don’t you end up with dozens of md files?

I’ve been copying and pasting the plan into the linear issue or PR to save it, but keep my codebase clean.

The biggest gaps are not in hardware or model size. There is a lot of logical fallacy in the industry. Most people believe bigger is better. For model size, compute, tools, etc.

The reality in ML is that small models can perform better at a narrow problem set than large ones.

The key is the narrow problem set. Opus can write you a poem, create a shopping list, and analyze your massive code base.

We trained our model to only focus on coding with our specific agent harness, tools, and context engine. And it’s small enough to fit on an M2 16GB. It’s as good as sonnet 4.5 and way better than qwen3.5:35b-a3b

Our beta will be out soon / rig.ai

AI has more training on web apps I think the real answer is that these guys are told they need to ship in 2 months and have a huge team of web devs

I feel like this is the silent majority. All the twitter hype is not representative of the real world.

Yeah but you're still using anthropic's subscription and tokens. That's not really an alternative. That's why we're shipping our own model with cortex.build

I am trying so hard to understand wtf people are excited about. I have failed. Claude Code can run over-night or while I'm out. Clawdbot looks like a great way to set tokens on fire.

ahem ... cortex.build

Current test version runs in 8GB @ 60tks. Lmk if you want to join our early tester group!

Yeah, I’ve hit this too. Once you do real agentic work or TDD, you’re optimizing context instead of code. That frustration is why we built Cortex: flat cost, no turn limits, runs locally, and git-aware context so you can just keep going. cortex.build

That's interesting. I use those moments to show it what not to do. Does it not just repeat the mistakes?

Can’t you use git for that? I do that often and just revert changes. It does require me to commit often but that’s probably good anyways.

I’ve never understood checkpoints / forks. When do you use them?

This is why we HAVE to have a local option and why we're building cortex.build. It's based on a small language model we trained exlusively for coding. We combine it with tools and a context graph designed to be more consistent than what's available today, especially with large codebases.

Tell me someone has turned this into a library we can use

What frustrates me is that Anthropic brags they built cowork in 10 days. They don’t show the seriousness or care required for a product that has access to my data.

We build our fine tuning and reinforcement pipeline at cortex.build by synthesizing interactions between a user, the agent loops, and a codebase. The exact data they get from users in Claude Code.

That data is critical to improve tool call use (both in correctness but also to improve when the agent chooses to use that tool). It's also important for the context rewrites Claude does. They rewrite your prompt and continuously manage the back-and-forth with the model. So does Cortex, just more aggressively with a more powerful context graph.

They are subsidizing Claude code so they can use your data to train better coding models. You’re paying them to show their models how to code better.

I think you just wrote a treatment for the next HBO Max sunday drama

There are lots of companies that do this. Doesn't make it right.

The real "evil" here is that companies like Meta, Google, and now OpenAI sell people a product or service that the customer thinks is the full transaction. I search with Google, they show me ads - that's the transaction. I pay for Chatgpt, it helps me understand XYZ - that's the transaction.

But it isn't. You give them your data and they sell it - that's the transaction. And that obscurity is not ethical in my opinion.

This. It’s the same play with their browser. They are building the most comprehensive data profile on their users and people are paying them to do it.

Proper sandboxing can fix this. But I didn’t see op mention it which I thought was weird

I agree with you. I don’t think number of startups or less reliance on funding is the measure though.

Businesses are not code. They solve problems, find their customers, convince them to buy their solution, and maintain that relationship.

Code has always been a factor but not the critical one.

Feel this so hard. The opposite is also true where you have a micro-service architecture and cursor faceplants in workspaces with multiple repos. We ended up building cortex.build partly because of this exact pain. Our context engine builds a git-aware dependency/provenance graph so it can stay local and only pull the relevant slice across a massive repo or dozens of smaller ones.

Context / understanding of your codebase is the most important part of AI coding. The current AI tools like claude code and cursor aren't super great at this on their own. They want you to use a plan mode to build up a clear plan of what files to touch and what needs done.

I like Augment Code for it's context engine - it is a lot better at finding where in the codebase changes need to be made without the planning.

It's the central thesis to what we're doing at Cortex.build - our local model can compete with large SOTA models because of how much work we do on mapping the codebase and creating just in time context for our LLM to make small and precise changes.

Perhaps it is optional for some. The point isn’t if it’s technically required. It’s that these items have less price sensitivity, which could be a greater factor in what industries invest in automation.

Sure but there’s a significant change in price that I can withstand before moving. My point is there’s less price sensitivity for some of these items. And because of that I wonder if that affects the amount of automation suppliers invest in.

Yeah you’re right, that’s fair. I just think people don’t behave that rationally. I’m not moving away from my kid’s grandparents because my local costs have gone up, for example.

There exists greater friction with many of the items in red than the highly automated ones.

My question is how linked is this friction to the lack of automation?

With text books and meat packing there are few players due to consolidation. This means they can avoid investing in automation and keep prices high because they face less resistance from consumers and virtually none from competitors.

In short I’m asking if market forces are to blame for lower automation. And therefore automation is not the root cause of price increases.