Rayline

The right model for every turn. We build ARC: open-weight routing intelligence for AI agents across cloud and on-device models.

[Rayline @RaylineAI]

ARC is out today!

[David Valerio Gilmore @davidvgilmore]

In 2015, I built an AI that detects early cancer.

Today, we’re introducing ARC by Rayline: the world’s first unified model for routing AI.

> 97.5% of Opus 5.5 quality
> Up to 1/10th of the cost
> Fully open-source

It picks the right model and the right amount of thinking for every step of your agent’s work.

Try it here and read our technical report below:

https://t.co/P7LCZwhMuQ

[David Valerio Gilmore @davidvgilmore]

In 2015, I built an AI that detects early cancer.

Today, we’re introducing ARC by Rayline: the world’s first unified model for routing AI.

> 97.5% of Opus 5.5 quality
> Up to 1/10th of the cost
> Fully open-source

It picks the right model and the right amount of thinking for every step of your agent’s work.

Try it here and read our technical report below:

https://t.co/P7LCZwhMuQ

[Eitan Rovero-Shein @tanchoco]

Stay tuned for Wednesday!

@RaylineAI

[chilang @chilang]

Wednesday 11/11

@RaylineAI

[Rayline @RaylineAI]

👀

[David Valerio Gilmore @davidvgilmore]

In 2015, I built an AI that detects early cancer.

It’s still running across hundreds of hospitals in the US today.

It was the most fulfilling work of my life, and I was building in AI years before everyone called it “AI”.

Building in AI then was a dream.

Now It feels like we’ve taken a turn for the worse.

It started when i was in undergrad.

I built my first company: AI models that made the political data transparent to voters.

At 21, I was offered six figures for it, but the buyers wanted to use it for reasons I couldn't ethically support.

So we turned down the offer and shut down the company instead.

In 2014, I moved into deep learning research on language models, for the Intelligence Community and Goldman Sachs.

Then HCA Healthcare came to us.

They put me in a competition against 10 people from IBM Watson. It was just me and the tools I’d built.

Our job was to build an AI model that can catch the cancer that slips through the cracks in hospitals.

It was the most intense 30 days of my life. Multiple all nighters.

My models won. That turned into a $5M deal, and later its own company, Azra AI.

Then the world caught on.

ChatGPT and Claude arrived, and AI costs exploded.

Building in AI in 2015 felt novel.

We were using it to find cancer tumours and save lives.

Now the biggest tech companies have turned AI into a money extraction machine.

That’s when I decided to make a change.

We raised $8.5M to build a better future for AI.

On Wednesday, we’re launching Rayline.

AI shouldn’t be locked inside a few labs’ models, prices and clouds. It should be open, transparent and decentralized.

More on Wednesday.

[Rayline @RaylineAI]

Repo: https://t.co/DudJVn5lyz

[Rayline @RaylineAI]

We open-sourced a local gateway for Claude Code.

The goal is simple: keep the Claude Code experience you already know, but use it with the models you want.

With the open-source gateway, you can connect Claude Code to other models using your own provider keys, whether that’s directly with model providers, through something like OpenRouter, or with local models running on your machine.

This is the more configurable version of what Rayline managed gives you out of the box: smarter model selection while preserving the coding-agent workflow you already like.

Rayline managed handles routing + inference for you.

The open-source gateway gives you control and lets you bring your own model connections.

GitHub repo link in thread.

[Example video below: Claude Code running with hybrid cloud + on-device AI. The main agent runs Opus in the cloud and orchestrates an Explore subagent that Rayline Local routes to a model running fully on device (Qwen3.6-35B-A3B, Q4), seamlessly, in a single session.]

[Rayline @RaylineAI]

There's a spectrum of model routing complexity.

Optimally routing hierarchical agents (with nested subagent tasks) across cloud and on-device is highly complex.

But will that be valuable?

David's article suggests that yes, it will be.

[David Valerio Gilmore @davidvgilmore]

https://t.co/oBdgJnQwCR

[Rayline @RaylineAI]

Model routing for coding agents must happen at the subagent point of delegation.

Doing so is the highest point of leverage for quality-neutral cost efficiency, and it will become more important as frontier models become better managers of deep agent hierarchies.

[David Valerio Gilmore @davidvgilmore]

https://t.co/6dHy5WPwaJ

[Rayline @RaylineAI]

@Arindam_1729 @davidvgilmore @steipete It's implemented as an AI gateway, yes. Just a one-link change and you can run Claude Code as a multi-model, multi-provider agent harness

[Rayline @RaylineAI]

Introducing Rayline: a model router built specifically for Claude Code.

Claude Code's API rates are 8-10x its subscription rates.

Most of those tokens are going to easy subtasks.

Plug in Rayline, and subagents get routed to open source and on-device models instead of burning Opus-level spend on grunt work.

Quality holds. Costs drop 60-90%.

What makes Rayline different:
- Routes at the subagent/subtask level
- On-device routing via MLX (Qwen 3.6 and others)
- Built-in ML router trained for Claude Code tasks
- Cloud fallback when Anthropic has outages
- Works with OpenAI models inside Claude Code

You keep using Claude Code exactly as you do today. Rayline handles the routing underneath.

We're already routing billions of tokens per day for individual developers and publicly traded companies.

Public beta is live now. Try it at Rayline[.]ai