Show HN: First Claude Code client for Ollama local models (opens in new tab)

(github.com)

44 pointsSerafimKorablev3mo ago27 comments

Just to clarify the background a bit. This project wasn’t planned as a big standalone release at first. On January 16, Ollama added support for an Anthropic-compatible API, and I was curious how far this could be pushed in practice. I decided to try plugging local Ollama models directly into a Claude Code-style workflow and see if it would actually work end to end.

Here is the release note from Ollama that made this possible: https://ollama.com/blog/claude

Technically, what I do is pretty straightforward:

- Detect which local models are available in Ollama.

- When internet access is unavailable, the client automatically switches to Ollama-backed local models instead of remote ones.

- From the user’s perspective, it is the same Claude Code flow, just backed by local inference.

In practice, the best-performing model so far has been qwen3-coder:30b. I also tested glm-4.7-flash, which was released very recently, but it struggles with reliably following tool-calling instructions, so it is not usable for this workflow yet.

27 comments

oceanplexian3mo ago

The Anthropic API was already supported by llama.cpp (The project Ollama ripped off and typically lags in features by 3-6 months), which already works perfectly fine with Claude Code by setting a simple environment variable.

davely3mo ago

Point of clarification: llama.cpp is MIT-licensed. Using it downstream (commercially or otherwise) is exactly what that license allows, so calling it a rip-off is misleading.

xd19363mo ago

And they reference that announcement and related information in the second line.

gcr3mo ago

Which announcement are you looking at? I see no references to llama-cpp in either Ollama's blog post or this project's github page.

d4rkp4ttern3mo ago

As others said this was possible for months already with llama-cop’s support for Anthropic messages API. You just need to set the ANTHROPIC_BASE_URL. The specific llama-server settings/flags were a pain to figure out and required some hunting, so I collected them in this guide to using CC with local models:

https://github.com/pchalasani/claude-code-tools/blob/main/do...

One tricky thing that took me a whole day to figure out is that using Claude Code in this setup was causing total network failures due to telemetry pings, so I had to set this env var to 1: CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC

eli3mo ago

There are already various proxies to translate between OpenAI-style models (local or otherwise) and an Anthropic endpoint that Claude Code can talk to. Is the advantage here just one less piece of infrastructure to worry about?

g4cg54g543mo ago

siderailing here - but got one that _actually_ works?

in particular i´d like to call claude-models - in openai-schema hosted by a reseller - with some proxy that offers anthropic format to my claude --- but it seems like nothing gets to fully line things up (double-translated tool names for example)

reseller is abacus.ai - tried BerriAI/litellm, musistudio/claude-code-router, ziozzang/claude2openai-proxy, 1rgs/claude-code-proxy, fuergaosi233/claude-code-proxy,

kristopolous3mo ago

What probably needs to exist is something like `llsed`.

The invocation would be like this

    llsed --host 0.0.0.0 --port 8080 --map_file claude_to_openai.json --server https://openrouter.ai/api

Where the json has something like

    { tag: ... from: ..., to: ..., params: ..., pre: ..., post: ...}

So if one call is two, you can call multiple in the pre or post or rearrange things accordingly.

This sounds like the proper separation of concerns here... probably

The pre/post should probably be json-rpc that get lazy loaded.

Writing that now. Let's do this: https://github.com/day50-dev/llsed

1 more reply

eli3mo ago

I've been hacking on this one for a few months now and it works for me https://github.com/elidickinson/claude-code-mux Been optimizing for routing to different models within one session so maybe overkill.

But I'm surprised litellm (and its wrappers) don't work for you and I wonder if there's something wrong with your provider or model. Which model were you using?

dsrtslnd233mo ago

What hardware are you running the 30b model on? I guess it needs at least 24GB VRAM for decent inference speeds.

derp-mcgee3mo ago

Im running qwen3-coder:30b-a3b-q8_0 @ 32k context. Comes out to 36gb and Im splitting it between a 3090 24gb and a 4060ti 16gb (ollama put 20gb on the 3090 and 13.5 on the 4060ti) , runs great tbh. Ollama running in ubuntu server and Im running claude code from my windows desktop pc.

thtmnisamnstr3mo ago

The general rule to follow is that you need as much VRAM as the model size. 30b models are usually around 19GB. So, most likely a GPU with 24GB of VRAM.

38362936483mo ago

But this also means tiny context windows. You can't fit gpt-oss:20b + more than a tiny file + instructions into 24GB

1 more reply

ryandrake3mo ago

I'd like to know this, too. I'm just getting started getting my feet wet with ollama and local models using just CPU, and it's obviously terribly slow (even 24 cores, 128GB DRAM. It's hard to gauge how much GPU money I'd need to plonk down to get acceptable performance for coding workflows.

storystarling3mo ago

I tried to build a similar local stack recently to save on API costs. In practice I found the hardware savings are a bit of a mirage for coding workflows. The local models hallucinate just enough that you end up spending more in lost time debugging than you would have paid for Sonnet or Opus to get it right the first time.

horacemorace3mo ago

I was trying to get Claude code to work with llama.cpp but could never figure out anything functional. It always insisted on a phone home login for first time setup. In cline I’m getting better results with glm-4.7-flash than with qwen3-coder:30b

g4cg54g543mo ago

~/.claude.json with {"hasCompletedOnboarding":true} is the key, then ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN work as expected

d4rkp4ttern3mo ago

Curious what llama-server flags you used. On my M1 Max 64GB MacBook I tried it in Claude Code (which has a 25K system message) and I get 3 tps.

But with Qwen3-30B-A3B I get 20 tps in CC.

dosinga3mo ago

this is cool. not sure it is the first claude code style coding agent that runs against Ollama models though. goose, opencode and others have been able to do that a while no?

d01003mo ago

Does this UI work with Open Code?

1 more reply

mchiang3mo ago

hey, thanks for sharing. I had to go to the Twitter feed to find the GitHub link:

https://github.com/21st-dev/1code

dang3mo ago

Thanks for catching that. I've changed the URL at the top to that from https://twitter.com/serafimcloud/status/2014266928853110862 now.

j / k navigate · click thread line to collapse

27 comments

oceanplexian3mo ago

davely3mo ago

Point of clarification: llama.cpp is MIT-licensed. Using it downstream (commercially or otherwise) is exactly what that license allows, so calling it a rip-off is misleading.

xd19363mo ago

And they reference that announcement and related information in the second line.

gcr3mo ago

Which announcement are you looking at? I see no references to llama-cpp in either Ollama's blog post or this project's github page.

d4rkp4ttern3mo ago

https://github.com/pchalasani/claude-code-tools/blob/main/do...

eli3mo ago

g4cg54g543mo ago

siderailing here - but got one that _actually_ works?

reseller is abacus.ai - tried BerriAI/litellm, musistudio/claude-code-router, ziozzang/claude2openai-proxy, 1rgs/claude-code-proxy, fuergaosi233/claude-code-proxy,

kristopolous3mo ago

What probably needs to exist is something like `llsed`.

The invocation would be like this

    llsed --host 0.0.0.0 --port 8080 --map_file claude_to_openai.json --server https://openrouter.ai/api

Where the json has something like

    { tag: ... from: ..., to: ..., params: ..., pre: ..., post: ...}

So if one call is two, you can call multiple in the pre or post or rearrange things accordingly.

This sounds like the proper separation of concerns here... probably

The pre/post should probably be json-rpc that get lazy loaded.

Writing that now. Let's do this: https://github.com/day50-dev/llsed

1 more reply

eli3mo ago

But I'm surprised litellm (and its wrappers) don't work for you and I wonder if there's something wrong with your provider or model. Which model were you using?

dsrtslnd233mo ago

What hardware are you running the 30b model on? I guess it needs at least 24GB VRAM for decent inference speeds.

derp-mcgee3mo ago

thtmnisamnstr3mo ago

The general rule to follow is that you need as much VRAM as the model size. 30b models are usually around 19GB. So, most likely a GPU with 24GB of VRAM.

38362936483mo ago

But this also means tiny context windows. You can't fit gpt-oss:20b + more than a tiny file + instructions into 24GB

1 more reply

ryandrake3mo ago

storystarling3mo ago

horacemorace3mo ago

g4cg54g543mo ago

~/.claude.json with {"hasCompletedOnboarding":true} is the key, then ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN work as expected

d4rkp4ttern3mo ago

Curious what llama-server flags you used. On my M1 Max 64GB MacBook I tried it in Claude Code (which has a 25K system message) and I get 3 tps.

But with Qwen3-30B-A3B I get 20 tps in CC.

dosinga3mo ago

this is cool. not sure it is the first claude code style coding agent that runs against Ollama models though. goose, opencode and others have been able to do that a while no?

d01003mo ago

Does this UI work with Open Code?

1 more reply

mchiang3mo ago

hey, thanks for sharing. I had to go to the Twitter feed to find the GitHub link:

https://github.com/21st-dev/1code

dang3mo ago

Thanks for catching that. I've changed the URL at the top to that from https://twitter.com/serafimcloud/status/2014266928853110862 now.

j / k navigate · click thread line to collapse