Pi-web is a browser interface to Pi coding agent, which can run on any server where you have Pi installed:

https://github.com/agegr/pi-web

Install pi-web, and it gives you a full-fledged UI, to control interactions with your Pi coding agent, all entirely in your web browser.

Installation is dead simple

Install and run pi-web at the command line of your application server:

PI_WEB_PASSWORD='your-password-123xyz!' \
PI_WEB_ALLOWED_HOSTS='pi.yourdomain.com' \
npx @agegr/pi-web@latest \
  --hostname 0.0.0.0 \
  --port 9247 \
  --no-open

If you want secure HTTPS, set up Caddy on your server:

pi.yourdomain.com {
  reverse_proxy localhost:9247
}

That's all it takes to make pi-web available at https://pi.yourdomain.com (or without encryption, at http://localhost.com:9247 (then you don't even need the Caddy piece). 1 minute and done.

Why use pi-web

One thing pi-web lets you do is use your own locally installed Pi agent like a ChatGPT/Claude/etc. interface, on any device that has a browser. Pi-web lets you select from any of the models you have configured in your local Pi installation (from Openrouter, Deepseek, OpenAI, Anthropic, Cline-pass, or any other LLM provider that Pi can be configured to use) - and that includes any locally hosted models running on any inference engine (LM Studio, Llama.cpp, Ollama, vLLM, etc.), which you've set up to run in Pi.

Pi-web uses all the same local configuration and session files as your existing Pi installation, so you can browse, resume, and export conversations, branch agent sessions, configure models and resources, inspect project files, etc., directly in your browser. And anything you do in your browser with pi-web, gets saved to your local Pi configuration, just as if you'd performed those operations in a command line session.

But that's just the beginning. You're not limited to running pi-web on a local computer, or using it just for chat. Install it on a VPS, so you can access Pi anywhere, and use all of Pi's deep agentic capabilities.

The main thing pi-web does is eliminate the need to use the command line of a local server, or SSH to access a remote server command line. Pi-web enables all the interactions you normally have with Pi at the command line, directly in your web browser. It provides a beautiful UI experience, suitable for users of any level. Your grandmother can run the pi-web user interface.

And of course, you're not limited to using Pi-web as a chat interface. Because pi-web is simply a web UI to Pi, you get all the same capabilities you'd get when running Pi at the command line of your PC, at the SSH command line of your VPS, etc. You can build software and run it immediately on your server, work with files on your hard drive, control the computer in any way you want...

This setup provides an extraordinarily flexible, complete environment, with a simple interface to any imaginable AI workflow, which can, for example, totally replace software development services such as Replit, Lovable, Bolt.net, etc., without any of the limitations or server expenses associated with those sorts of commercial providers. Install any infrastructure, software packages, libraries, and tools you want, run self-hosted local models to process private data, and maintain compliance requirements in any way you need. You have complete control, because you run everything on your own machine(s). Pi-web simply gives users access to the full command line capabilities of a local Pi agent, without ever needing to touch a command line.

Adding Pi-web's easy web interface to your server forms the foundation of an end-to-end self-hosted AI production environment, which can be run fully from any device that has a modern browser.

Rent a VPS for $5/month, subscribe to Cline-pass for $7/month, or use any other LLM provider, including your own self-hosted models, and you've got an entire production environment where you can build and deploy software of any complexity, perform research, manage any hardware - do anything you can do with any computing system - all controlled entirely from any client device that runs a browser (any form factor works, any OS, old hardware works, etc.).

Get a VPS account from Contabo (or OVH Cloud, or any other competitor) for ~$5 per month. Them get an account from Cline-pass for ~$80 per year (less than $7 per month). If you use the newest Flash models like Deepseek v4.1, GLM 5.3, and Mimo 2.6, you'll get virtually endless monthly use of near-frontier capable models, without the constant rate limiting or pay-per-token fees of other LLM providers - and lots of access to even the biggest open-source models (Kimi K3, Qwen 3.8 Max, GLM 5.3, Mimo 2.6 Pro, etc.). Add a domain name from any Registrar, and this entire setup costs less that $150 per year.

If you want to run a fully self-hosted AI system, from the ground up, get an old laptop with a mobile RTX 3080ti, or any other GPU which is able to run a model such as Qwen 3.6 35a3, and you've got a significantly capable, fully off-grid, private setup, without any rate limits whatsoever, for a one time cost of around $1000 total. Of course you can spend tens of thousands of dollars on GPU equipment to run the best open source LLMs, but a little Qwen model can accomplish a lot of useful daily work. The point is, pi-web can make the whole setup much simpler to use.

I first set up pi-web after purchasing a couple old Amazon Fire devices at a garage sale for $5, and wanted a way to access my own AI models on them. Pi-web worked absolutely beautifully on those cheap little machines (which have huge operational OS limitations), and also worked immediately on my phone, old Chromebooks, desktop PCs, tablets, old netbooks and Linux boxes, etc., all without installing any software on any of those machines. You just need to open any browser to your pi-web URL, enter your password, and the entire environment is available and fully capable of doing anything you'd normally have to be at a computer console to control. You can move between locations and access a single production development environment which contains all your files, using any device you have available at any site.

This is a cheap, simple, light weight solution to serious AI capabilities which imposes virtually no limitations, and is flexible enough to incorporate any LLM provider you want to set up.

I've got a few clients now who build their own software prototypes using Pi on VPS accounts. They currently log into their VPS server over SSH, and run Pi at the server command line. I've given them all a few hours of training to do that, and some basic prompt examples to use in Pi, to create software with Flask & SQLite, to set up separate Python environments running in separate tmux sessions, with each application configured to run on separate ports, to perform regular backups, etc., and I've provided a basic agents.md file which keep those users from unintentionally shutting down processes on the server, deleting files, etc., to prevent likely issues from casual prompting missteps which non-technical don't know to avoid.

Pi-web now opens up that sort of environment and workflow to users who are so non-technical that they don't even want to dive into using a command line, SSH, or any other core tools. They can now simply open their browser, start a conversation just like they do with ChatGPT, request applications to be built & run, and those apps are instantly available on their VPS where pi-web is running. To those users, that is absolute magic. The browser interface removes virtually every bit of resistance to getting started. Non-technical users can immediately begin performing development and debug iterations like a pro. Pi-web makes it all so familiar and easy - just like using ChatGPT. Showing users how to use Pi and other agents at a command line then also turns into a much simpler task in the future, because all the pieces are already familiar.

It takes me about 10-15 minutes to set up an entire pi-web instance, once a VPS account and an LLM provider account (OpenRouter, Cline-pass, etc.) have been set up. I can typically walk clients through the pre-requisite hosting account setup steps in about 1/2 hour. So, in less than an hour, a full private AI infrastructure can be configured, from the ground up.

What an absolutely awesome solution pi-web is, to so many sorts of workflows. It has the ability to make computing so much easier, in general, for the masses: a simple web based UI to a simple natural language AI agent that can be run on virtually any client device, which requires virtually zero training, to enable the completion of virtually any computing task, of nearly any complexity. That's a big deal.

Pi-web is a browser interface to Pi coding agent, which can run on any server where you have Pi installed:

https://github.com/agegr/pi-web

Install pi-web, and it gives you a full-fledged UI, to control interactions with your Pi coding agent, all entirely in your web browser.

Installation is dead simple

Install and run pi-web at the command line of your application server:

PI_WEB_PASSWORD='your-password-123xyz!' \
PI_WEB_ALLOWED_HOSTS='pi.yourdomain.com' \
npx @agegr/pi-web@latest \
  --hostname 0.0.0.0 \
  --port 9247 \
  --no-open

Set up Caddy on your server:

pi.yourdomain.com {
  reverse_proxy localhost:9247
}

That's all it takes to make pi-web available at https://pi.yourdomain.com (or at http://localhost.com:9247 - then you don't even need the Caddy part)!

Why use pi-web

One thing pi-web lets you do is use your own locally installed Pi agent like a ChatGPT/Claude/etc. interface, on any device that has a browser. Pi-web lets you select from any of the models you have configured in your local Pi installation (from Openrouter, Deepseek, OpenAI, Anthropic, Cline-pass, or any other LLM provider that Pi can be configured to use) - and that includes any locally hosted models running on any inference engine (LM Studio, Llama.cpp, Ollama, vLLM, etc.), which you've set up to run in Pi.

Pi-web uses all the same local configuration and session files as your existing Pi installation, so you can browse, resume, and export conversations, branch agent sessions, configure models and resources, inspect project files, etc., directly in your browser. And anything you do in your browser with pi-web, gets saved to your local Pi configuration, just as if you'd performed those operations in a command line session.

But that's just the beginning. You're not limited to running pi-web on a local computer, or using it just for chat. Install it on a VPS, so you can access Pi anywhere, and use all of Pi's deep agentic capabilities.

The main thing pi-web does is eliminate the need to use the command line of a local server, or SSH to access a remote server command line. Pi-web enables all the interactions you normally have with Pi at the command line, directly in your web browser. It provides a beautiful UI experience, suitable for users of any level. Your grandmother can run the pi-web user interface.

And of course, you're not limited to using Pi-web as a chat interface. Because pi-web is simply a web UI to Pi, you get all the same capabilities you'd get when running Pi at the command line of your PC, at the SSH command line of your VPS, etc. You can build software and run it immediately on your server, work with files on your hard drive, control the computer in any way you want...

This setup provides an extraordinarily flexible, complete environment, and interface to any imaginable AI workflow, which can, for example, completely replace software development services such as Replit, Lovable, Bolt.net, etc., without any of the limitations and server expenses associated with those sorts of commercial providers. Install any infrastructure, software packages, libraries, and tools you want, run self-hosted local models to process private data, and maintain compliance requirements in any way you need. You have complete control, because you run everything on your own machine(s). Pi-web simply gives users access to the full command line capabilities of a local Pi agent, without ever needing to touch a command line.

Adding Pi-web's simple web interface to your server forms the foundation of a fantastically useful, end-to-end self-hosted AI production environment, which can be run fully from any device that has a modern browser.

Rent a VPS for $5/month, subscribe to Cline-pass for $7/month, or use any other LLM provider, including your own self-hosted models, and you've got an entire production environment where you can build and deploy software of any complexity, perform research, manage any hardware - do anything you can do with any computing system - all controlled entirely from any client device that runs a browser (any form factor works, any OS, old hardware works, etc.).

For less than $7 per month on Cline-pass, if you use the newest Flash models like Deepseek v4.1, GLM 5.3, and Mimo 2.6, you get virtually endless monthly use of near-frontier capable models, without the constant rate limiting or pay-per-token fees of other LLM providers - and lots of access to even the biggest open-source models (Kimi K3, Qwen 3.8 Max, GLM 5.3, Mimo 2.6 Pro, etc.). Add $5 per month for a VPS at Contabo, and a domain name from any Registrar, and this entire setup costs less that $150 per year.

If you want to run a completely self-hosted AI system, from the ground up, get an old laptop with a mobile RTX 3080ti, or any other GPU which is able to run a model such as Qwen 3.6 35a3, and you've got a significantly capable, fully off-grid, private setup, without any rate limits whatsoever, for around $1000 total.

I first set up pi-web after purchasing a couple old Amazon Fire devices at a garage sale for $5, and wanted a way to access my own AI models. Pi-web worked absolutely beautifully on those cheap little machines (which have huge operational OS limitations), and also worked immediately on my phone, Chromebooks, desktop PCs, tablets, old Linux boxes, etc., all without installing any software on any of those machines. You just need to open a browser to your pi-web URL, enter your password, and the entire environment is available and fully capable of doing anything you'd normally have to be at a computer console to control. You can move between locations and access a single production development environment which contains all your files, using any device you have available at any site.

This is a cheap, simple, light weight solution to serious AI capabilities which imposes virtually no limitations, and is flexible enough to incorporate any LLM provider you want to set up.

I've got a few clients now who build their own software prototypes using Pi on VPS accounts. They currently log into their VPS server over SSH, and run Pi at the server command line. I've given them all a few hours of training to do that, and some basic prompt examples to use in Pi, to create software with Flask & SQLite, to set up separate Python environments running in separate tmux sessions, with each application configured to run on separate ports, to perform regular backups, etc., and I've provided a basic agents.md file which keep those users from unintentionally shutting down processes on the server, deleting files, etc., to prevent likely issues from casual prompting missteps which non-technical users don't know enough to avoid.

Pi-web now opens up that sort of environment and workflow to users who are so non-technical that they don't even want to dive into using a command line, SSH, or any other typical core tools. They can now simply open their browser, start a conversation just like they do with ChatGPT, request applications to be built & run, and those apps are instantly available on their VPS where pi-web is running. To those users, that is absolutely magic, and the browser interface removes virtually every bit of resistance to getting started. Non-technical users can immediately begin performing development and debug iterations like a pro. Pi-web makes it all so familiar and easy - just like using ChatGPT. Showing users how to use Pi and other agents at a command line then also turns into a much simpler task in the future, because all the pieces are already familiar.

It takes me about 10-15 minutes to set up an entire pi-web instance, once a VPS account and an LLM provider account (OpenRouter, Cline-pass, etc.) have been set up. I can typically walk clients through all those account setups in about 1/2 hour.

What an absolutely awesome solution this is, to so many sorts of workflows, and to making computing so much easier in general for the masses: a simple web based UI to a simple natural language AI agent that can be run on virtually any client device, with virtually zero training, to complete virtually any computing task of nearly any complexity. That's a big deal.

Pi-web is a browser interface to Pi coding agent, which can run on any server where you have Pi installed:

https://github.com/agegr/pi-web

You install pi-web, and it gives you a full-fledged UI to control your interactions with pi coding agent, all entirely in your web browser.

Installation is dead simple

Install and run pi-web at the command line of your application server:

PI_WEB_PASSWORD='your-password-123xyz!' \
PI_WEB_ALLOWED_HOSTS='pi.yourdomain.com' \
npx @agegr/pi-web@latest \
  --hostname 0.0.0.0 \
  --port 9247 \
  --no-open

Set up Caddy on your server:

pi.yourdomain.com {
  reverse_proxy localhost:9247
}

That's all it takes to make pi-web available at https://pi.yourdomain.com!

Why use pi-web

One thing pi-web lets you do is use your own locally installed Pi agent like a ChatGPT/Claude/etc. interface, on any device that has a browser. Pi-web lets you select from any of the models you have configured in your local Pi installation (from Openrouter, Deepseek, OpenAI, Anthropic, Cline-pass, or any other LLM provider that Pi can be configured to use) - and that includes any locally hosted models running on any inference engine (LM Studio, Llama.cpp, Ollama, vLLM, etc.), which you've set up to run in Pi.

Pi-web uses all the same local configuration and session files as your existing Pi installation, so you can browse, resume, and export conversations, run agent turns, configure models and resources, inspect project files, etc., directly in your browser. And anything you do in your browser with pi-web, gets saved to your local Pi configuration, just as if you'd performed those operations in a command line session.

But that's just the beginning. You're not limited to running pi-web on a local computer, or using it just for chat. Install it on a VPS, so you can access Pi anywhere, and use all of Pi's deep agentic capabilities.

The main thing pi-web does is eliminate the need to use SSH to access your server command line. Pi-web provides all the interactions you normally have with Pi at the command line, directly in your web browser. It provides a beautiful UI experience suitable for users of any level. Your grandmother can run this user interface.

And of course, you're not limited to using Pi-web as a chat interface. Because pi-web is simply a web UI to Pi, you get all the same capabilities you'd get when running Pi at the command line of your PC, at the SSH command line of your VPS, etc. You can build software and run it immediately on your server, work with files on your hard drive, control the computer in any way you want...

This setup provides an extraordinarily flexible, complete environment & interface to any imaginable AI workflow, which could, for example, completely replace software development services such as Replit, Lovable, Bolt.net, etc., without any of the limitations and server expenses associated with those sorts of commercial providers. Install any packages and libraries you want, run self-hosted local models to process private data, and maintain compliance requirements in any way you need. You have complete control, because you run everything on your own machine(s). Pi-web simply gives users access to the full command line capabilities of a local Pi agent, without ever needing to touch a command line.

Adding Pi-web's simple web interface to your server forms the foundation of a fantastically useful, end-to-end self-hosted AI production environment, which can be run fully from any device that has a modern browser.

Rent a VPS for $5/month, subscribe to Cline-pass for $7/month, or use any other LLM provider, including your own self-hosted models, and you've got an entire production environment where you can build and deploy software of any complexity, perform research, manage any software - do anything you can do with any computing system - all controlled entirely from any client device that runs a browser (any form factor works, any OS, old hardware works, etc.).

For less than $7 per month on Cline-pass, if you use the newest Flash models like Deepseek v4.1, GLM 5.3, and Mimo 2.6, you get virtually endless use of near-frontier capable models - and lots of access to all the other big open-source models (Kimi K3, Qwen 3.8 Max, GLM 5.3, Mimo 2.6 Pro, etc.). Add $5 per month for a VPS at Contabo, and a domain name from any Registrar, and this entire setup costs less that $150 per year.

If you want to run a completely self-hosted AI system, get an old laptop with a mobile RTX 3080ti, or any other GPU which is able to run a model such as Qwen 3.6 35a3, and you've got a significantly capable, fully off-grid, private setup.

I set up pi-web after purchasing a few old Amazon Fire devices at a garage sale for $5, and wanted a way to access my own AI models. Pi-web worked absolutely beautifully on those cheap little machines (which have huge operational OS limitations), and it works immediately on my phone, Chromebook, desktop PC, tablet, etc., all without installing any software on any of those machines. Just open the browser at your pi-web URL, enter your password, and the entire environment is available and fully capable of doing anything you'd normally have to be at a computer console to control.

This is a cheap, simple, light weight solution to serious AI capabilities which imposes virtually no limitations, and is flexible enough to incorporate any LLM provider you want to set up.

I've got a few clients now who build their own software prototypes using Pi on VPS accounts. They currently log in over SSH, and run Pi at the server command line. I've given them all a few hours of training, and basic prompt examples to use in Pi, to create software with Flask & SQLite, to use separate Python environments running in separate tmux sessions on separate ports, to perform regular backups, etc., and I've provided a basic agents.md file which keep those users from unintentionally shutting down processes on the server, deleting files, etc., to prevent likely issues due to casual prompting missteps which non-technical users don't know enough to avoid.

Pi-web now opens up that sort of environment and workflow to users who are so non-technical that they don't want to use the command line, SSH, or any other typical tools. They can now simply open their browser, start a conversation just like they do with ChatGPT, request applications to be built & run, and those apps are instantly available on their VPS where pi-web is running. To those users, this is absolutely magic, and there's virtually no resistance to getting started. Those users can perform development and debug iterations like a pro. It's all so familiar and easy - just like using ChatGPT. And then showing those users how to use Pi at the command line, is much easier in the future, because all the pieces are already familiar.

It takes me about 10-15 minutes to set up the whole thing, once a VPS account and an LLM provider account (OpenRouter, Cline-pass, etc.) have been set up. I can typically walk clients through all those account setups in about 1/2 hour.

What an absolutely awesome solution to so many workflows, and to making computing so much easier in general for the masses: a simple web based UI to a simple natural language AI agent that can be run on virtually any device, to complete virtually any computing task.

Pi-web is a browser interface to Pi coding agent, which can run on any server where you have Pi installed:

https://github.com/agegr/pi-web

You install pi-web, and it gives you a full-fledged UI to control your interactions with pi coding agent, all entirely in your web browser.

Installation is dead simple

Install and run pi-web at the command line of your application server:

PI_WEB_PASSWORD='your-password123xyz!' \
PI_WEB_ALLOWED_HOSTS='pi.yourdomain.com' \
npx @agegr/pi-web@latest \
  --hostname 0.0.0.0 \
  --port 9247 \
  --no-open

Set up Caddy on your server:

pi.yourdomain.com {
  reverse_proxy localhost:9247
}

That's all it takes to make pi-web available at https://pi.yourdomain.com

Why use pi-web

One thing pi-web lets you do is use your own locally installed Pi agent like a ChatGPT/Claude/etc. interface, on any device that has a browser. Pi-web lets you select from any of the models you have configured in your local Pi installation (from Openrouter, Deepseek, OpenAI, Anthropic, Cline-pass, or any other LLM provider that Pi can be configured to use) - and that includes any locally hosted models running on any inference engine (LM Studio, Llama.cpp, Ollama, vLLM, etc.), which you've set up to run in Pi.

Pi Web uses all the same local configuration and session files as your existing Pi installation, so you can browse and resume conversations, run agent turns, configure models and resources, and inspect project files directly in your browser.

But that's just the beginning. You're not limited to running pi-web on a local computer, or using it just for chat. Install it on a VPS, so you can access Pi anywhere, and use all its agentic capabilities.

With pi-web, you don't need SSH access to your server command line. Pi-web provides all the interactions you normally have with Pi at the command line, directly in your web browser. It provides a beautiful UI experience suitable for users of any experience level. Your grandmother can run this user interface.

And of course, you're not limited to using Pi-web as a chat interface. Because pi-web is simply a web UI to Pi, you get all the same capabilities you'd get when running Pi at the command line of your PC, at the SSH command line of your VPS, etc. You can build software and run it immediately on your server, work with files on your hard drive, control the computer in any way you want...

This setup provides a super flexible complete environment and interface to any imaginable AI workflow, which could for example replace services such as Replit, Lovable, Bolt.net, etc., without any of the limitations and server expenses associated with those sorts of commercial providers. Install any packages and libraries you want, run self-hosted local models to process private data, and maintain compliance requirements in any way you need. You have complete control, because you run everything on your own machine(s). Pi-web gives users access to the full command line capabilities of a local Pi agent, without ever needing to touch a command line.

This forms the foundation of an absolutely killer end-to-end, complete self-hosted AI production environment, which can be run fully from any device with a modern browser.

Rent a VPS for $5/month, subscribe to Cline-pass for $7/month, or use any other LLM provider, including your own self-hosted models, and you've got an entire production environment where you can build and deploy software of any complexity, perform research, manage any software - do anything you can do with any computing system - all controlled entirely from any device that runs a browser.

For less than $7 per month on Cline-pass, if you use the newest Flash models like Deepseek v4.1, GLM 5.3, and Mimo 2.6, you get virtually endless use of near-frontier capable models - and lots of access to all the other big open-source models (Kimi K3, Qwen 3.8 Max, GLM 5.3, Mimo 2.6 Pro, etc.). With an additional $5 per month for a VPS at Contabo, and a domain name from any Registrar, this entire setup costs less that $150 per year.

I set up pi-web after purchasing a few old Amazon Fire devices at a garage sale for $5, and wanted a way to access my own AI models. It worked so beautifully on those cheap little machines (which have all sorts of usage limitations), and it works on my phone, Chromebook, desktop PC, tablet, etc., without installing anything on any machine. Just open the browser at your pi-web URL, enter your password, and the entire environment is available and fully capable of doing anything you'd normally have to be at a computer console to control.

This is a cheap, simple, light weight solution to serious AI capabilities which impose virtually no limitations. I've got a few clients now who build their own software prototypes using Pi on VPS accounts. They currently log in over SSH, and run Pi at the server command line. I've given them all a few hours of training, and basic prompts for Pi to use Flask & SQLite, to use separate Python environments running in separate tmux sessions, etc., and I've provided a basic agents.md file which keep those users from unintentionally shutting down processes on the server, deleting files, performing regular backups, etc., to avoid issues which could be experienced with just some casual prompt missteps that non-technical users aren't likely to think of.

Pi-web now opens up this sort of environment and workflow to users who are so non-technical that they don't want to use the command line, SSH, or any other typical tools. They can now simply open their browser, start a conversation just like with ChatGPT, request applications to be built & run, and the apps are instantly available on their VPS where pi-web is running. To those users, this is absolutely magic, and there's virtually no resistance. They can perform iterations like a pro. It's so familiar and easy - just like using ChatGPT.

It takes me about 10-15 minutes to set up the whole thing, once a VPS account and an LLM provider account (OpenRouter, Cline-pass, etc.) have been set up. I can typically walk clients through all those account setups in about 1/2 hour.

What a freakin awesome solution.

Pi-web is a browser interface to Pi coding agent, which can run on any server where you have Pi installed:

https://github.com/agegr/pi-web

You install pi-web, and it gives you a full-fledged UI to control your interactions with pi coding agent, all entirely in your web browser.

Installation is dead simple

Install and run pi-web at the command line of your application server:

PI_WEB_PASSWORD='your-password123xyz!' \
PI_WEB_ALLOWED_HOSTS='pi.yourdomain.com' \
npx @agegr/pi-web@latest \
  --hostname 0.0.0.0 \
  --port 9247 \
  --no-open

Set up Caddy on your server:

` pi.yourdomain.com { reverse_proxy localhost:9247 }

Now pi-web is available at https://pi.yourdomain.com

Why use pi-web

One thing pi-web lets you do is use your own locally installed Pi agent like a ChatGPT interface, on any device that has a browser. Pi-web lets you select from any of the models you have configured in your local Pi installation (from Openrouter, Deepseek, OpenAI, Anthropic, Cline-pass, or any other LLM provider that Pi can be configured to use) - that includes any locally hosted models running on any inference engine (LM Studio, Llama.cpp, Ollama, vLLM, etc.), which you've set up to run in Pi.

Pi Web uses all the same local configuration and session files as your existing Pi installation, so you can browse and resume conversations, run agent turns, configure models and resources, and inspect project files directly in your browser.

But that's just the beginning. You're not limited to running pi-web on a local computer, or using it just for chat. Install it on a VPS, so you can access Pi anywhere, and use all its agentic capabilities.

With pi-web, you don't need SSH access to your server command line. Pi-web provides all the interactions you can have with Pi, directly in your web browser, with a beautiful UI experience suitable for users of any experience level. Your grandmother can run this UI.

And of course, you're not limited to using Pi-web as a chat interface. Because pi-web is simply a web interface to Pi, you get all the same capabilities you'd get when running Pi at the command line of your PC, at the SSH command line of your VPS, etc. You can build software and run it immediately on your server, work with files on your hard drive, control the computer in any way you want, etc.

This setup provides a super flexible AI interface, which could for example replace services such as Replit, Lovable, Bolt.net, etc., without any of the limitations and server expenses associated with those sorts of commercial providers. Install any packages and libraries you want, run privately hosted models to process private data, and maintain compliance requirements in any way you need. You have complete control, because you're running everything on your own machine(s).

This forms the foundation of an absolutely killer end-to-end, complete self-hosted AI production environment, which can be run from any device with a modern browser. Rent a VPS for $5/month, subscribe to Cline-pass for $7/month or use any other LLM provider including your own self-hosted model, and you've got an entire production environment where you can build and deploy software of any complexity, perform research, manage any software - do anything you can do with a computing system - all controlled entirely from any device that runs a browser.

For less than $7 per month on Cline-pass, if you use the newest Flash models like Deepseek v4.1, GLM 5.3, and Mimo 2.6, you get virtually endless use of near-frontier capable models - and lots of access to all the other big open-source model (Kimi K3, Qwen 3.8 Max, GLM 5.3, Mimo 2.6 Pro, etc.). With an additional $5 for a VPS at Contabo, and a domain name from any Registrar, this entire setup costs less that $150 per year.

I set up pi-web because I purchased a few old Amazon Fire devices at a garage sale for $5 this weekend, and wanted a way to access my own AI models. It works beautifully on those cheap little machines which have all sorts of usage limitations, and it works on any phone, Chromebook, desktop PC, tablet, etc.

This is a cheap, simple, light weight solution to serious AI capability that imposes virtually no limitations. I've got a few clients now who build their own software prototypes using Pi on VPS accounts. They currently log in over SSH, and run Pi at the server command line. I've given them all a few hours of training, and basic prompts for Pi to use Flask, SQLite, separate Python environments running in separate tmux sessions, etc., and I've provided a basic agents.md file which keep them from unintentionally shutting down processes on the server, deleting files unintentionally, etc.

Pi-web opens up this sort of environment to users who are non-technical enough to not want to use the command line, SSH, etc. They can now open their browser, start a conversation, request application to be built/run, and the apps are instantly available on their VPS where pi-web is running. To those users, this is absolutely magic, and there's virtually no resistance. It's just like using ChatGPT. It takes me about 10-15 minutes to set up the whole thing, once a VPS account and LLM provider account (OpenRouter, Cline-pass, etc.) has been set up. I can typically walk clients through all those account setups in about 1/2 hour.

What a freakin awesome solution.