Introducing TogetherLink! An open source CLI to run any open source model inside your favorite coding harness. Run GLM 5.2 directly in Codex and Claude Code.

Try TogetherLink here to run open source models in Codex and Claude Code!

TogetherLink is also fully open source! Shoutout to @riccardogiorato for building this with me.

GLM 5.2 in Codex is big. The benchmark gap vs closed models is shrinking. This is the "bring your own model" moment for AI coding.

Great idea, hope I can use more open weight models.

Please lmk what you think if you do!

Awesome! will be using this stat

Please lmk what you think!

Hey, I love what you just built, it's amazing! But if you're open to some feedback, next time you record a video, maybe put your personal video on top or in a better spot. That way, viewers can see what's at the bottom of your screen, which would really help with visibility and just make for a better video overall.

the harness lock-in was always the real friction, not the model. dropping GLM 5.2 behind the same agent loop is the part that actually matters. do you get tool-calling parity across models or is that still on the user?

Model-agnostic tooling is quietly becoming the portable layer. Your workflow and habits carry over; the model becomes a swappable part. That inversion — process over provider, is where the leverage sits now.

Curious how togetherlink plays out at scale

oh this is neat. how are you handling tools that differ between Codex and Claude Code? is TogetherLink translating a shared tool schema, or mostly swapping the model endpoint underneath?

Open source is amazing 🚀

Running GLM 5.2 straight inside Claude Code or Codex is a genuinely useful bridge

@nutlope inside Claude Code do you still pay Claude token usage for any reasoning/compute/agent session in order to route to/use open source models?

This is amazing, I can't wait to try it! ✨

Is it a proxy or are you also deploying the model to cloud provider of users choice?

nice and clean, love it!

thanks!

🚀🚀

🙏🙏🙏

Open source models sound fun 🤔

this lands because the harness was always the hard part. running glm inside claude code means you inherit the loop, the tools, the retries. swapping the model is a config line. rebuilding that scaffolding is a month.

This is the kind of tooling that makes open models actually usable in production. Being able to swap GLM 5.2 or Ling into Codex without changing your workflow changes the cost calculus overnight.

This is very cool! What happens when you route a message to the sub-agent.. do you see their thinking etc or just waiting for the response and its delivered back in your default CLI?

essentially this is just something which changes the env for a claude code or a codex session so that it connects to an open source model right? or does it have something else too? for different models right now just use oc/kilo

Yeah so this will set configs + ships with a proxy to translate between chat completions and the OpenAI responses API + Anthropic messages API. Can definitely use opencode for different models, this is just a way to use CC/Codex if you really want to!

Running open source models in Claude Code? This is a dream come true, babe! 🥹 You're a lifesaver. Testing this right now! 🚀✨

Will my Opencode sub work with it?

Not currently, it requires a Together AI key. But it’s open source so I’m sure you can hack it to accept that!

I was looking for something like this lately but can we get something where I can use my supergrok subscription in codex

cool but the bottleneck isn't which model runs where. it's context routing. agent loops break at 40-60k when tool output floods the window. running open models locally is a flex. what's the actual task you'd route to GLM vs Claude?

少切窗口这点挺实用

nice work! i built something similar with