You can now run Claude Code locally for FREE. → Using open-source models → Locally and fully private → Full agent + tool workflows Here's the FULL step-by-step tutorial to get it all set up in less than 10 minutes:

✍️Best practices: - Make sure to have the latest versions of Ollama and Claude Code - Set the context length in Ollama settings of at least 32K - Use glm-4.7, gpt-oss, qwen3-coder for >= 16gb vram (smaller models might not be too useful for complex tasks) - Expect slower responses than the cloud API - Keep Ollama running in the background before launching Claude Code

This isn’t Claude sir, it’s an Anthropic-compatible interface wired to Ollama 🍿🥃

Claude Code = the tool (still Anthropic's). Claude Opus/Sonnet = the model. The tutorial swaps the model, not the tool. You're literally running Claude Code, just with a different LLM backend via Ollama's Anthropic-compatible API

Running the Claude Code binary doesn’t mean you’re running Claude. You’re running a client that expects Claude, pointed at a different model. That’s not the same thing sir.

@dr_cintas you are right. situation is clear. this is an android phone in an iphone box.

@dr_cintas Good article, great hack. The title is a bit clickbaity, but technically it’s solid.

that's is cool ah but I ain't letting ollama touch my code brother 😅😅 opus is king

But isn’t using the Claude models one of the biggest benefits of Claude Code?

@ollama Which open-source models have people found to be the best for coding?

Glad I followed you 😃. This is exactly what I need but for #claudebot by @steipete. -Any advice/vid please?

@steipete Maybe it’s in the works 👀

@steipete Wow you made my day! -definitely worth the wait and thanks in advance 😎

That was easy! I went with qwen3-coder

Great share! This makes it so easy to run local models using Claude Code. AI is making our lives so easy.

How successful is that with tool calling?

We tried, but anyway you need a paid subscription. Without that it won't allow you to login in Claude code or CLI. @AnthropicAI and @ollama integration are good but still need $17 monthly subscription. A kind of Freemium, where @claudeai rate limits, @ollama helps.

@gizmothegizzer know we were talking about this then other week, thought you might find it interesting!

Nice one!

Posted here as well 👀

it's just insane running to check out the guide

Local agents are underrated. No API latency. No usage anxiety. Full control over tools + memory. This is how serious builders experiment

this is huge for developers in regions where API costs add up fast. running locally with Ollama means teams in the Gulf, South Asia, and Africa can experiment with agent workflows without burning through credits. tested this setup with qwen3-coder on a MacBook and it handles most file editing and refactoring tasks surprisingly well. the latency tradeoff is worth it when you're iterating on personal projects or learning the tool.

A more easy way

That's the feature I needed to use it.

now you just need to do a clawdbot tutorial to make it work locally as well

Even with 32 gb ram mine was so slow and rich 22gb of ram usage and never answer to a simple « hello how are you » 😅 I tried with gpt-oss and qwen3-coder

That's cool. We design UIs that iterate fast.

Very nice, thanks!

tried this setup last week. performance is obviously slower than api but for experimentation and learning how the tool workflows actually work its perfect. also nice knowing nothing leaves your machine when youre prototyping with sensitive codebases

Nice! Local Claude Code could be a game changer for private agent workflows.

what is use case of this? 20b model for coding?

Or you can just use this script

That's awesome! Running Claude Code locally is a game-changer for privacy.

so what if the url is xai-grok and a api-key

u can use cc switch as easier to build it

@grok kan du förklara allt?

at this point, use clawdbot.

I try same and got : API Error: 404 404 page not found

Amazing