# [How C3] Cursor, VScode or Claude Code If you want to use an open source SOTA model to vibecode you can go with Kimi K2 which has 1000b params or you can also go with Qwen3 Coder 480b, both work just fine. Comput3 is hosting Kimi on 8xb200s and Qwen3 on 6xb200s so that you can access these massive models at q8 with full context super easy with just your comput3 API key. Lemme explain how you can install claude code and set things up with your comput3 acc. ## Cursor ### Step 1: Install Cursor Download and install [cursor](https://cursor.com/) on your system. The installation is straight forward. ### First Option: Cursor Pro User You can natively integrate your comput3 models into the cursor agent system very easy. Its the simplest integration that you'll find in this tutorial ![image](https://hackmd.io/_uploads/ryuDr9Khlg.png) Navigate to the cursor settings and open Models. ![image](https://hackmd.io/_uploads/SkrxHqK2xg.png) Open the API Keys section and flip the switch that says Override OpenAI Base URL. ![image](https://hackmd.io/_uploads/B1srBcY3ee.png) Copy and paste the comput3 api endpoint there `https://api.comput3.ai/v1` and then insert your API key Now flip the other switch to activate your OpenAI API Key. ![image](https://hackmd.io/_uploads/BysJ89thll.png) Confirm the popup to enable the connection with comput3. ![image](https://hackmd.io/_uploads/SJjGIcF3lx.png) Click View All Models and then hit Add Custom Model ![image](https://hackmd.io/_uploads/SJkKLqK3eg.png) Type kimi-k2 and hit Add. Scroll down to find more models that you can use with your comput3 account. ![image](https://hackmd.io/_uploads/By0t85Knxg.png) ![image](https://hackmd.io/_uploads/rkVx_qY2ge.png) Thats it you're ready to go to vibecode with kimi or any other model thats provided by comput3. This is a snapshot of whats hosted in September 2025. ![image](https://hackmd.io/_uploads/HJVavcYnxl.png) ### Start Vibecoding Set up once you can now activate the model in your chat window. From here its same as always. ![image](https://hackmd.io/_uploads/HJmadcKnxl.png) With the model of your choice enabled you can start to prompt cursor. ![image](https://hackmd.io/_uploads/BJ3Sd9Knxg.png) You're good to go. Join the [official discord](https://discord.gg/EFZPGm9qUu) and follow the comput3 on twitter and keep track about the latest models that are being added into the system. ![image](https://hackmd.io/_uploads/S1Tft5F2gg.png) ### Second Cursor Option: Install Kilo Plugin Even without a cursor pro subscription you can use comput3 models via the Kilo extension. I personally prefer the native cursor experience tho and would recommend it to anyone interesting in using cursor with comput3 to go that route. But feel free to go ahead and try this route. ![image](https://hackmd.io/_uploads/HJJY9fsKgl.png) Hit Ctrl+Shift+X to open the extension manager. Search and install Kilo. ![image](https://hackmd.io/_uploads/SJ5h9zoFxx.png) ### Start Vibecoding Launch Kilo and click Use your own API key. ![image](https://hackmd.io/_uploads/BJNIozjYxe.png) You can now launch Kilo from the sidebar. Select OpenAI Compatible, set the comput3 base url `https://api.comput3.ai/v1` and get your api key from the comput3 dashboard and paste it in. Then select Kimi k2. Scroll down a bit, set 128k context window, disable image support and enable prompt caching. ![image](https://hackmd.io/_uploads/BJYnjzstlx.png) You can also select `deepseek-v3.1` or `qwen3-coder:480b` if you want. the context window of qwen3 is 256k context since the latest update. Switch between the models depending on the current task but kimi-k2 is a great place to start. [<img src="https://hackmd.io/_uploads/BkrOnMoKee.jpg" width="49%" style="padding-right:1%"/>](https://hackmd.io/_uploads/BkrOnMoKee.jpg)[<img src="https://hackmd.io/_uploads/Syji2fiYgl.jpg" width="49%" style="padding-left:1%"/>](https://hackmd.io/_uploads/Syji2fiYgl.jpg) When working with the agent you can toggle between multiple specialists or modes. There is Architect to design a concept and Code for software development. Now try to one shot a simple game. Send a simple prompt `create a simple tetris game with sophisticated UX UI like what you'd expect from any successful mobile/web game from the 2020s be creative when it comes down to the aesthetics and make sure not to use any image files at all. everything has to be rendered on load. we need this game written in vuejs make sure to use most bleeding edge wegbl webgpu and threejs tools and innovation` and see what the agent is doing. ![image](https://hackmd.io/_uploads/rJSyafoFge.png) Pro tip: Hit the arrow in the top left sidebar. Pin kilo there and move it to the top using drag and drop to ensure quick access to your agents. ## Visual Studio Code One of the best Vibecoding Setups out there. The Compatibility with Open Source models and mcps is amazing. This is a sonnet4 grade vibecoding IDE that you setup in 10 mins. ![image](https://hackmd.io/_uploads/SkOxMDDFel.png) ### Step 1: Install vscode Download and install [vscode](https://code.visualstudio.com/) on your system. The installation is straight forward. ### Step 2: Install Kilo Plugin ![image](https://hackmd.io/_uploads/HkQM1vvFxg.png) Get the free plugin from the [official vscode market](https://marketplace.visualstudio.com/items?itemName=kilocode.kilo-code) and copy the launch command from the website or here: `ext install kilocode.Kilo-Code` and paste it into the top bar in your vscode to install it within minutes. ![image](https://hackmd.io/_uploads/rJhiJPDtxl.png) ### Start Vibecoding You can now launch Kilo from the sidebar. Select OpenAI Compatible, set the comput3 base url `https://api.comput3.ai/v1` and get your api key from the comput3 dashboard and paste it in. Then select Kimi k2. Scroll down a bit, set 128k context window, disable image support and enable prompt caching. [<img src="https://hackmd.io/_uploads/B1yzlwPFxl.jpg" width="49%" style="padding-right:1%"/>](https://hackmd.io/_uploads/B1yzlwPFxl.jpg) You can also select `deepseek-v3.1` or `qwen3-coder:480b` if you want. the context window of qwen3 is 256k context since the latest update. Switch between the models depending on the current task but kimi-k2 is a great place to start. [<img src="https://hackmd.io/_uploads/rkBQZwDKex.jpg" width="49%" style="padding-right:1%"/>](https://hackmd.io/_uploads/rkBQZwDKex.jpg)[<img src="https://hackmd.io/_uploads/SkIrbDvtxx.jpg" width="49%" style="padding-left:1%"/>](https://hackmd.io/_uploads/SkIrbDvtxx.jpg) When working with the agent you can toggle between multiple specialists or modes. There is Architect to design a concept and Code for software development. Now try to one shot a simple game. Send a simple prompt `create a simple minigolf game with sophisticated UX UI like what you'd expect from any successful mobile/web game from the 2020s be creative when it comes down to the aesthetics and make sure not to use any image files at all. everything has to be rendered on load. we need this game written in vuejs make sure to use most bleeding edge wegbl webgpu and threejs tools and innovation` and see what the agent is doing. ## Claude Code Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows -- all through natural language commands. Use it in your terminal, IDE, or tag @claude on Github. [<img src="https://hackmd.io/_uploads/r1kZ0zmKge.png"/>](https://hackmd.io/@reneil1337/c3-vibe) ### Step 1: Install Claude Code (only initially) *Prerequisites: You'll need npm to install claude code, simply follow this [guide](https://docs.npmjs.com/downloading-and-installing-node-js-and-npm) on how to download and install things according to your OS. Once thats done you can go ahead.* Launch your Terminal and install Claude Code via this command: `npm install -g @anthropic-ai/claude-code` ![image](https://hackmd.io/_uploads/HJm-4zQKxl.png) *You only need to do this step once. The other two steps need to be executed everytime you open a terminal to launch claude code.* ### Step 2: Set some Anthropic variables Paste the following 3 commands into your terminal. `export ANTHROPIC_BASE_URL=https://api.comput3.ai` Make sure to put in your own comput3 API key otherwise it will not work. `export ANTHROPIC_AUTH_TOKEN=your-comput3-api-key` Last one depending on the model you wanna use: `export ANTHROPIC_MODEL=kimi-k2` or `export ANTHROPIC_MODEL=qwen3-coder:480b` ![image](https://hackmd.io/_uploads/B1qqVGmFxg.png) *Keep in mind that always have to export these variables before launching claude in step 3.* ### Step 3: Launch Claude Code One final command in your terminal and you're greeted by Kimi K2 (or Qwen3 Coder) from the comput3 API instead of Sonnet 4. Ready to vibe? Enjoy! `claude` ![image](https://hackmd.io/_uploads/BypxBzXYxe.png) ![image](https://hackmd.io/_uploads/B1sMrM7Fgx.png) ![image](https://hackmd.io/_uploads/r1kZ0zmKge.png) ## Other comput3 Tutorials Do more with your account. Visit and bookmark this continously growing [c3 tutorial page](https://hackmd.io/@reneil1337/comput3) that explains all sorts of things your comput3 account enables. From your own personal ChatGPT to agentic systems and media generation theres tons of stuff to explore. ## How to Find your Comput3 API Key After you've [acquired](https://dexscreener.com/solana/34qhkhrhyningwruftqjnb2vfv8oqyqd5tvdwwde1man) and [staked](https://app.streamflow.finance/staking/solana/mainnet/DGZebyog1twdFaKwN2KZkBLxDkzX4rh7L61S6TAty8et) some $com token you can [login to your dashboard](https://launch.comput3.ai/) with your solana wallet and grab your API key at the bottom page. You can do way more things in this dashboard like spin up comfyui instances or launch dedicated ollama instances with broader model selection but in this tutorial we keep it simple. ![image](https://hackmd.io/_uploads/BJGO16mreg.png) If you run into issues join the [official comput3 Discord server](https://discord.gg/EFZPGm9qUu) and ask the community.