NCA CLI: This Rust AI Coding Agent is 10x Lighter Than the Competition 🦀

Tony Xhepa| 00:08:22|May 5, 2026

Chapters9

Introduction to the native coding agent NCA CLI and its lightweight design.

NCA CLI is a Rust-native coding agent that runs as a tiny single binary, using far less RAM than cloud agents while offering easy terminal setup and OpenRouter/OpenAI options.

Summary

Tony Xhepa walks through NCA CLI, a native Rust-based coding agent that runs as a single binary in the terminal. He emphasizes its tiny memory footprint—claims of under 20 MB of RAM and roughly 10x memory savings versus cloud code agents like OpenCode. The video demonstrates installation steps from nca-cli.com, plus a contrast between running OpenAI/OpenRouter in a local environment versus the NCA native agent. Tony shows memory comparisons using Warp and Alacritty terminals, noting NCA itself uses as little as 8 MB during a simple interaction, compared with OpenCode hovering around hundreds of MB. He walks through a sample setup with OpenRouter, including creating an API key and configuring the nca config file in the ~/.nca directory, and highlights supported providers (Minimax, OpenAI, Anthropic, OpenRouter) with how to select models. The tutorial also covers how to export the API key and set the base URL for OpenRouter, plus hints about seeing other agents like Pi that are heavier in memory. The takeaway is that NCA offers a native, lightweight alternative for coding tasks right in the terminal, with practical setup steps and provider choices outlined for free usage through OpenRouter and other providers.

Key Takeaways

NCA CLI claims under 20 MB RAM and around 10x memory savings versus cloud coding agents.
The video demonstrates real-world memory comparisons: OpenCode ~400 MB, Warp terminal ~300 MB, Alacritty ~42 MB, versus NCA at about 8 MB during use.
Setup requires visiting nca-cli.com, choosing a provider (OpenRouter, Minimax, OpenAI, Anthropic), and exporting an API key to config the NCA directory (~/.nca).
OpenRouter is highlighted as a free model option, with an API key and model like OWL (free for 1,000,000 input tokens, per the video).
Configuration includes creating an API key on OpenRouter, exporting it in the shell, and setting the base URL to https://openrouter.ai/api/model in the NCA config.

Who Is This For?

Developers curious about lightweight, locally-run AI coding assistants and Rust-based tools who want to minimize memory usage while still accessing popular AI providers.

Notable Quotes

"NCA is Ras native single binary terminal first orchestrate coding agents from the terminal interactive 2i"

—Tony introduces NCA as a native, single-binary terminal agent.

"NCA is under 20 mgaby of RAM on the order of 10 times lighter than the cloud code or open code"

—He emphasizes the memory advantage over cloud-based agents.

"the NCA is down there. Where is the NCA? It's hard to find"

—A moment showing how NCA runs unobtrusively in the UI.

"NCA is only 8 mgabytes"

—Demonstrating the actual RAM footprint during use.

"open router API key which is going to be able with the key you are going to create here"

—Describes configuring the OpenRouter provider with an API key.

Questions This Video Answers

How does NCA CLI achieve such low memory usage compared to OpenAI or OpenCode?
What are the exact steps to set up NCA CLI with OpenRouter in a terminal like Warp or Alacritty?
Can I run multiple AI providers (OpenRouter, OpenAI, Anthropic) in NCA CLI at the same time?
What are the minimum system requirements to run NCA CLI on MacOS or Linux?
Is NCA CLI suitable for large-scale coding sessions or just quick prompts in the terminal?

NCA CLIRust OpenRouter OpenAIMinimaxAnthropicAlacrittyWarp terminalSingle binaryRAM efficiency

Full Transcript

Hello friends, Tony here. Welcome. In today's video, I'm going to show you a new coding agent which is this native coding engine nca-cli.com is Ras native single binary terminal first orchestrate coding agents from the terminal interactive 2i and you can see is under 20 mgaby of RAM on the order of 10 times lighter than the cloud code or open code far less memory than most competing agents. I'm going to open I have installed. Now to install you just go to this website nca-cli.com and here we have the install install. Copy this one and is available for Mac OS able silicon Intel and Linux also. Okay. We can go also the documentation and read more. We have also providers here. But for now, let's just me I have opened the activity monitor. I'm going to open let's open the warp terminal. Okay. And yeah, here is warp terminal is almost 300 mgabytes. And also I'm going to open the alacrity. Alacity is uh where is alacrity is not here. Yeah, it's down there is only 42 mgabytes and for that I have downloaded the alacrity when I want to save some memory I use. Now let's let's use also only the if I run if I navigate the herd example. Okay. And open the open code here is the open code. Now you can see we have open code here which is 400 mgabytes. If we run something else, just let's say hi here. Yeah, as you can see, 400 megabytes. And now we have the response. How can I help you? Okay, now let's see. Let's close this and let's open this uh native CLI agent I have installed now I'm going to run just NCA and here I'm using the open router and the model is all alpha but just let's see here where is the NCA we have Camtasia which I'm I'm recording this uh screen but we have Google Chrome also spotlight rcast Fast activity monitor. The NCA is down there. Where is the NCA? It's hard to find. We can search here. Yeah, NCA is only 8 mgabytes. Okay. And let's say for example just hi here also okay hello how can I help you but again is only 8 mgabytes and for that if you wanted to use to save some RAM and to use this NCA native coding agent go and download and here we have providers on the documentation we have providers and supported providers have minimax anthropic openi and open router we don't have uh local ls here. But yeah, if you want for free to use for free, you can use the open router. Go to open router.ai create an account and then here go to preferences and create an API key. After that, so here you can create an API key. Give it a name and then go to models. You can search for free models. Okay? or I'm using this OWL which for now is uh free as you can see $0 for the million input tokens and also zero for output tokens. Just copy this. Now I'm going to show you how to set up. You also can find it right here. So you need to create in the NCAA directory the config to file. This is the configuration file used for the NCA. Let's me just exit from here. So slashzit and I'm going to navigate the cd.nca and I'm going to open this with VS Code. And here this is the configuration to file provider default we have the open router you can find it here also. So this is an example with minimax provider anthropic openai open router you can have here also if you go to configuration this is for all the providers what I have here I have only for minimax openi and open router but yeah you can copy here and you have minimax open router anthropic and openai now default I have choser And we have the provider for minimax open AI and then open router. Here you need to say API key and which is this open router API key and you need to export this on your in my example is in the SSRC export this open router API key which is going to be able with the key you are going to create here. Okay, I'm going to show you in a minute. Then the base URL is https open router.ai/appi model. Copy this and paste it here. Or you can choose what model you like. And yeah, optionally if you want to add also this here. This is how to set up with open router. Now let me just show you also envy. And I'm going to zoom it this little bit. And vim src. And here I have export open router_appi_key eel with the key here. And in this way you have the NCA ready to work with the open router any mode that you like. Okay friends and that's it all about this video. What I wanted to show you a new native code agent. This is Ras native is uh only 20 megabytes of RAM 10 times lighter than the cloud code or open code. So far less memory than most competing agent. And just because we are here, let's open I'm going to see I'm going to open the pi agent because also the pi is light. Let's see now where is the pi agent. Here is also the pi agent as you can see is 95. Yeah, this one NCA is far less memor memory than most competing agents. Now, if you like such a videos, don't forget to subscribe to my channel, like the video, share with your friends, and I'm going to see you in another one. All the best.