Skip to content
BitPull.cloud
ProductNetworkShared codePlansEarnOur vision
Log inAccountDownload
Language
Live network

Your computers and your friends’, pooled for AI

Join your computers and your friends’ to run large AI models none of you could run alone, and earn when you are not using them. Works with your favourite AI tools.

Download the appTry it free, no card needed

Free app for macOS, Windows and Linux. Models on your own computers use 0 credits.

BitPullQwen 3.5 4B
ChatProjectsComputersEarn

Add a function that sums a list, with a test

Here it is, with a test:

export function sum(xs: number[]) {
  return xs.reduce((a, b) => a + b, 0);
}

test("sum", () =>
  expect(sum([1, 2, 3])).toBe(6));
Apply to sum.ts✓ sum.test.ts · 1/1

Free · on your computers

Group model on 3 computers

  1. The model is split across the group's computers
  2. Each question passes through all of them
  3. The answer comes back as if from one big computer

One app for chat, code and your computers

Chat with open and paid models, build sites and code with an agent, and see what each of your computers is doing. See every feature

Computers9
  • MacBook ProM3 Pro · 18 GBRunning
  • Office PCRTX 4090 · 24 GBIn the group
  • Gaming PCRTX 3090 · 24 GBIn the group
  • Mac miniM4 Pro · 48 GBLending
  • Home laptopRyzen 7 · 16 GBIdle
  • ThinkPad X1Core Ultra 7 · 32 GBLending
  • Ubuntu 24.04EPYC · 128 GBIn the group
  • Windows 11RTX A5000 · 24 GBLending
  • iMacM3 · 24 GBOffline
Llama 3.1 70BYour group · 3 computers
MacBook Pro1–27
Office PC28–54
Gaming PC55–80

Can my computers run Llama 3.1 70B?

Not one alone: the model needs about 40 GB of memory. MacBook Pro, Office PC and Gaming PC have 66 GB together, so your group runs it.

Free · on your group's computers

Now run the tests and build the site on the Office PC

Done. It ran in a sandbox on the Office PC, deleted when the job ended:

vitest  ✓ 214 passed  12.4 s
vite    ✓ built in 3.1 s · dist/ 412 kB

Write a message

No bigger computer needed

There is knowledge in every project and potential in computers sitting idle. We are building BitPull to connect them, and make more possible together.

Four ways people use BitPull

BitPull is not only for programmers. Pick what you came for: each page explains it in its own words.

Build a website or an app

Describe it in your own words. The AI builds it, shows it to you and puts it online. No programming needed.

$ claude mcp add bitpull \
    -- npx -y bitpull mcp
✓ bitpull

Code faster, spend less on AI

MCP tools for Claude Code, Codex and Cursor, open models on your own GPUs, sandboxed jobs and an editor with an agent.

+3.16

Earn with the computer you don't use

Lend it while it is idle. Every job earns you credits, which you can spend on AI or withdraw as euros.

Put computers together with friends

Classmates, friends or a team join their computers to run large AI models none of them could run alone.

One large model across several computers

Friends, classmates or a team join a group with an invitation. A model too big for any of their computers runs on all of them at once.

  1. The model is split across the group's computers
  2. Each question passes through all of them
  3. The answer comes back as if from one big computer
See how groups work
QuestionLlama 3.1 70B
MacBook ProLayers 1–27
Office PCLayers 28–54
Gaming PCLayers 55–80
Answer66 GB

Your idle computer earns credits

Lend it when it is idle, earn credits on every job and spend them on any model, or withdraw the ones you earned as euros. See what lending pays

22:00 – 07:006 × CPU · 16 GB

Lends only when you say

Lending is off until you turn it on. Pick the hours, cores and memory it may use.

+3.16RTX 4090 · 1 h

Covers its electricity

Prices are set so the credits cover the power your computer uses while it works for others.

SandboxDeleted at the end

Keeps your files out of reach

Jobs run in a sandbox with no access to your files, deleted when the job ends.

Withdraw

Pays out in euros

Earned credits mature in 14 days. Withdraw them as euros from €20.

Paid plans, from €4.50 a month

Cheaper than Runpod for the same work: a 24 GB GPU from €0.13 an hour, an RTX 4090 €0.25, with prices that cover the electricity of whoever lends.

Compare the plans
Pay as you goFor occasional use: you pay only for what you use.
€0
BaseFor regular use of open models and small groups.
€4.50/month
ProFor heavy use: more credits, larger groups and any model.
€9.90/month
MaxFor the heaviest use, for less than a typical AI subscription.
€19.90/month

Prices include VAT. Recharge any amount from €5, at 16 credits per euro. Plan credits work on open models; paid models use the credits you buy. Bought and plan credits are for your use on BitPull; credits earned by lending can be withdrawn as euros.

Works with Claude Code, Codex, Cursor and Gemini CLI

They connect through MCP with one command: npx -y bitpull mcp

  • Claude Code
  • Codex
  • Cursor
  • Gemini CLI
MCP
  • Your computersOpen models, free
  • The networkBigger GPUs, paid in credits
  • Shared codeCode others shared, read file by file

Frequently asked questions

What is BitPull?
BitPull is a free desktop app (macOS, Windows, Linux) for anyone who works with AI: build websites by describing them, use AI models for less, earn by lending a computer you are not using, or put computers together with friends. Developers connect it to Claude Code, Codex and Cursor through MCP to reuse shared code and computing power. Using other people's resources costs credits; lending yours earns them.
Is BitPull only for developers?
No. People use it to build websites by describing them, to chat with AI models for less, to earn by lending a computer they are not using, and to put computers together with friends to run larger models. Developers also connect it to Claude Code, Codex or Cursor. bitpull.cloud/for/creators, /for/developers, /for/lenders and /for/groups explain each one.
Do I need to know how to program to use BitPull?
No. To build a website you describe it in your own words in Projects (“New site”), watch it take shape in the preview and ask for changes; the AI writes the code. To lend your computer or use AI models in the chat you do not write any code either. Programmers get extra tools, but nobody has to use them.
How does it help when my AI credits run out?
Your AI finds code that others have already written, signed and described, and reads only the files it needs, which can save tokens. Simple tasks such as summaries, tests or routine changes can go to open AI models running on community computers.
What do I need to install?
Only the BitPull app to use the network. To lend CPU and memory you also need Docker or Apple Container; to lend AI models, Ollama or the image, speech and music engine that BitPull installs by itself. The app shows what is required and what is optional on your computer.
How do credits work?
You spend credits when you use other people's resources and earn 80% of what others spend on yours. With pay as you go there is no monthly fee and you recharge any amount from €5. Base (€4.50 a month), Pro (€9.90) and Max (€19.90) include monthly credits for open models, larger groups and priority.
How much does BitPull cost compared with Runpod and GPU clouds?
Less for the same work, even when you pay with a recharge (1 credit ≈ €0.0625). An hour of an RTX A5000-class card with 9 cores and 25 GB costs 2.15 credits (≈ €0.13) against $0.16 on Runpod; an RTX 4090 3.95 credits (≈ €0.25) against $0.34; data-centre cards about 10% less than Runpod (an H100 about 24.6 credits, ≈ €1.54, against $1.99); a million tokens of a medium open model 23 credits (≈ €1.44) against $10 for Qwen3 32B. Only the RTX 3090, which draws much more power, costs about as much as on Runpod. Runpod rents data-centre machines with guaranteed availability; BitPull uses computers people lend, which may be busy, and the computer that runs a job can read it.
Can I earn money by lending my computer?
A little, and that is not the point. Prices are set so the credits you earn cover the electricity your computer uses while it works for others, counted at Italian household prices (among the dearest in Europe), with a little left over. Withdrawn as euros a credit is worth less (0.02 € today), so with a gaming graphics card the euros usually do not cover the electricity; with the processor only, or a data-centre card, something is left. The real advantage is sharing: the computer you already have and the electricity you already pay become credits you spend on other people's computers (larger models, more power when yours is busy) instead of paying a subscription to a big company. Example: an RTX 4090 working for others for an hour uses about 0.55 kWh (≈ €0.17) and earns 3.16 credits, worth about €0.20 spent on BitPull.
Can my AI assistant install and set it up?
Yes. AI assistants find everything at bitpull.cloud/llms.txt and bitpull.cloud/install.json, with verified installers and the list of required and optional components. BitPull is in the official MCP registry: "npx -y bitpull mcp" connects it to Claude Code, Codex, Cursor and others, and you approve the login on the website.

Start with the computers you have

Install the app, connect your AI tool, and add a friend's computer when you need more.

Product names and logos are trademarks of their owners.

Download the app

01Product

One app. Everything connected.

Write code, talk to models, use the terminal and pick up from your phone. Explore what you can do with BitPull.

BitPull
sum.tsREADME.md
1import { expect, test } from "vitest";
2
3export const sum = (xs: number[]) =>
4  xs.reduce((a, b) => a + b, 0);
5
6test("sum", () => expect(sum([1, 2, 3])).toBe(6));
Chatqwen2.5-coder · 7B
Add a function that sums a list, with a test

Here it is, with a test:

export const sum = (xs: number[]) =>
  xs.reduce((a, b) => a + b, 0);

test("sum", () => expect(sum([1, 2, 3])).toBe(6));
Apply to sum.ts
↑
TerminalFree · on your computers
$ 
✓ sum.ts (1 test) · 0.4s
Chat with your modelsExamples with sample data.
A code editor like Cursor

Projects has VS Code's file icons, extensions from Open VSX (themes, icons, snippets, languages), integrations the AI agent can use, and Run and Debug with breakpoints and variables.

Quick commands and shortcuts in Projects and the chat

Type / for commands like /compact, /review or /init and your own custom commands, ! to run a command in the project's folder, @ for a file and ? for the keyboard shortcuts, which you can remap. Drag a folder onto the app to work on it.

Open models in your terminal

Type bitpull chat kimi to talk to a model like you would with claude, or bitpull code to let the agent work on the folder you are in, asking you before each change. Other programs, like aider, can use the same models.

Dark like Cursor, chat like claude.ai

From version 0.1.37 the app starts in a dark theme with Cursor's colours, or light, or as your system. The chat has a side bar with search, projects, pinned and recent chats, and a composer with attachments, your commands, integrations and dictation, on a phone too.

Talk to it, and group your chats

From version 0.1.38 the microphone in the chat dictates your message and shows clearly when it is on; the waveform button starts a spoken conversation, with the answers read aloud. Speech becomes text with Whisper on your own computer when it is installed (free, the voice stays there), otherwise with a model of the network. Chats can be pinned and put in named groups that fold open, and groups can be pinned too.

Scheduled tasks, set by any AI

From version 0.1.38 the chat has a Scheduled tab, as in claude.ai: a reminder at a time, or a prompt that a model answers every day, on weekdays, every week or once, with the answer saved as a new chat. The app's chat, your AI tools through MCP, and anything in a terminal with `bitpull schedule` can add and change them. They run on your computer while the app is open.

Build a site with AI, watch it, publish it

From version 0.1.41 Projects has a preview beside the editor: the agent starts your site, sees the page, reads its errors, clicks buttons and fills forms while you watch, on a computer, tablet or phone size. “New site” makes a ready folder, and Publish puts it online on your own Cloudflare or Vercel account and creates its database (Supabase, Neon) or Redis (Upstash), on your account or free and temporary without one. Your keys stay on your computer. The same preview works from a terminal and for Claude Code, Cursor and other AI tools through MCP. The chat shows the pages a model writes (HTML, SVG) working beside the conversation, as artifacts do.

Control your computers from your phone

Open bitpull.cloud/app on an iPhone or Android phone, scan the code shown by the app on your computer, and use that computer's Projects, terminal and chat from the phone, with screens made for a phone. Everything runs on your computer; the link between phone and computer is encrypted so that the BitPull server cannot read it. Off until you turn it on, and only your own computers.

10:42BitPullOfficeHome$ npm testrunning…✓ 412 passed38.2s · office-pc$ python train.pyepoch 3/10RTX 4090 · home-pc✓Office▶ 3/10Home
  1. 01From the café, you type on your phone
  2. 02The command runs on your office computer, the result comes back
  3. 03Then on the one at home. The link is encrypted

Examples with sample data.

Frequently asked questions

Can I use BitPull and follow my work from my phone?

Yes. Open bitpull.cloud/app on any phone (iPhone, Android and others, no app store needed). In the BitPull app on your computer turn on “Use from your phone” and scan the code it shows: from then on the phone opens that computer's Projects (files, editor, terminal with the keys a phone lacks, AI assistant, Git), chat and settings. Commands always run on your computer; the link between phone and computer is encrypted and the BitPull server cannot read it. It works only with your own computers, which must be on with BitPull open.

Does Projects have slash commands and keyboard shortcuts?

Yes, from version 0.1.32. In Projects and in the chat, / opens a menu of commands (summarise the conversation, review the changes, write the project's instruction file, choose the model and more) plus your own custom commands written as Markdown files. In Projects, ! runs a command in the project's folder and the AI answers about its output. Keyboard shortcuts include a queue that waits while the AI works, history, Ctrl+X then Ctrl+S to send the queue at once, and readline editing; every shortcut can be changed in a file.

Can I use BitPull's AI models from the terminal?

Yes, from version 0.1.33. bitpull models lists the models you can use; bitpull chat followed by a name (for example bitpull chat kimi) opens a conversation in the terminal, and bitpull code runs the Projects agent on the current folder, showing every change and command for you to approve. In the app's own terminal the bitpull command is already there; elsewhere install it with npm install -g bitpull. Programs that speak the OpenAI API, like aider, can use the same models through the app.

Does BitPull's editor have extensions and a debugger?

Yes, from version 0.1.36. Projects installs extensions from Open VSX, the open registry Cursor uses: colour themes, file icon themes, snippets and language grammars (their code is never run). Integrations from the official MCP registry give the AI agent tools such as GitHub or Atlassian, each use approved by you. Run and Debug starts Node.js and Python programs with breakpoints, variables, watch expressions, the call stack and a debug console, with the same keys as VS Code.

Can I remove a folder from BitPull without deleting it?

Yes, from version 0.1.37. In Projects, the folder menu has "Remove a folder's authorisation": BitPull stops seeing that folder, while the folder and its files stay on your computer as they are. "Open a new project" in the same menu authorises another folder.

Can an AI remind me of something, or run a prompt at a set time?

Yes, from version 0.1.38. In the chat, Scheduled lists reminders (a notification, free) and prompts a model answers at a time (the answer becomes a new chat; a paid model costs credits each time), once, every hour, every day, on weekdays or every week. You can make them yourself, ask the chat ("remind me every Monday at 9 to send the report"), or let an AI tool add them through MCP or with `bitpull schedule` in a terminal. They run on your computer while the BitPull app is open; one missed while it was closed runs when it opens, marked late.

Can I talk to BitPull's models by voice?

Yes, from version 0.1.38. In the chat, the microphone dictates into the message box, and the waveform button opens a spoken conversation: you speak, the model answers and the answer is read aloud with your system's voice; tap the circle to interrupt. Speech becomes text with Whisper on your computer when it is installed (free; install it in the control panel under the AI models), otherwise with a speech model of the network, paid in credits. It works in the desktop app and from your phone; the first time, the system asks to allow the microphone.

Can BitPull build and publish a website for me, like Lovable?

Yes, from version 0.1.41. In Projects choose “New site”, describe it, and the agent builds it, checking it in the preview beside the editor. Then press Publish: the site goes online on your own Cloudflare or Vercel account (both have free plans), and with one click it gets a database (Supabase or Neon) or Redis (Upstash) whose keys go to the project's .env.local. You connect each account once with a key you create on its site; for Neon and Upstash you can also start without an account, with a free database that lasts 72 hours unless you claim it. The site and the keys go from your computer straight to those services, never through BitPull's server. The agent can be one of your own models (free) or a paid one; it always asks before publishing.

Can my AI (Claude Code, Cursor, the agent in BitPull) see and test the site it is building?

Yes, from version 0.1.41. The preview shows sites served on your computer (localhost). The AI reads the page as text with numbered buttons, links and fields, gets its console errors and failed requests, clicks, types, scrolls and takes screenshots; you see each click happen in the preview. The agent in Projects and `bitpull code` use it directly; Claude Code, Cursor and other MCP clients use the tools preview_open, preview_look, preview_click, preview_type, preview_scroll and preview_screenshot while the BitPull app is open. With a model of the community, each click or text waits for your approval. Only local sites can be previewed, never sites on the internet.

BitPull

Build with what we have. Together.

Start with your computers. Connect your AI. Make room for what comes next.

Download the app↗Try it free, no card needed↗

02Network

Shared computing

Builds, tests, data processing and open AI models run on community computers, in sandboxes deleted when the job ends. Your data stays encrypted until it reaches the computer that runs the job.

Claude Code · Codex · Cursor● bitpull.run_job("npm test")⎿ 3 computers · sandbox✓ 412 passed · 38.2srelayMacBook M336 GBsandbox ✓RTX 409024 GB VRAMsandbox ✓Linux server64 GB RAMsandbox ✓
  1. 01Your AI sends a job
  2. 02BitPull runs it on free computers, in a sandbox
  3. 03The result comes back to your AI

Shared groups: put your computers together

Classmates or friends join a group with an invitation and each offers their own computers, to run large open AI models none of them could run alone. Inside the group the models are free; traffic is paid in credits, and lending the group to others earns them back. Use a group only with people you trust.

BitPull
My networkgroup · 3
Mac miniM2 · 16 GBonline
PCRTX 3090 · 24 GBonline
LaptopRyzen · 32 GBjoined with an invitationonline
Memory together40 GB
llama-3.3-70b · Q4 · 42 GB ▾Start the group model
Loading the model on 3 computers…Running · your group's questions go first
Your computers working togetherExamples with sample data.

How it works

Download the app↗
1

Install the app

On macOS, Windows or Linux. Lending stays off until you turn it on.

2

Connect your AI tool

Claude Code, Codex, Cursor and others reach BitPull through MCP.

3

Jobs run on the network

On your own computers first, then on the community's, in isolated sandboxes.

4

Save, or earn by lending

Pay only for what you use, and earn credits when others use your computer.

Is it safe?

  • Our server cannot read your jobs: they stay encrypted until they reach the computer that runs them.
  • Jobs run without internet and without access to the lender's files.
  • You decide when, how much and what to lend.

How security works

Private data? Use only your own computers. A computer that runs a job can read what is in it. For anything private choose "Only my computers": it never reaches anyone else, and it is free.
After a job, your data is deleted Your data travels encrypted and is opened only inside an isolated space on the computer that runs the job. BitPull gives the person lending no tool to see it. As soon as the result is delivered, it is deleted from that computer, which does not keep the key to read it again; the encrypted copies in transit are deleted for good within 24 hours at most. If you lend: the job cannot read your files or use the internet, and leaves nothing on your computer. If the app or the computer stops in the middle of a job, what it left is removed before the next one starts.

For lenders

For lenders: earn credits with your idle computer

Lending is off until you turn it on. You choose when and how much, you earn 80% of what each job costs, and jobs have no access to your files.

Explore↗
BitPull
LendingOff until you turn it onOn
✓Only when the computer is idle
Max CPU50%
Max memory8 GB
job 7f3asandbox · no internet · no access to your files✓ done · sandbox deleted
job c91esandbox · no internet · no access to your files✓ done · sandbox deleted
Credits earned+0
What you lend, and whenExamples with sample data.

Frequently asked questions

Can several people put their computers together to run a large AI model?

Yes, from version 0.1.29. In the app's panel, Shared groups: one person creates a group and sends the invitation; each member offers the computers they want. A copy of a model uses up to 9 computers over the internet or 24 on the same network, and more computers give more copies, so more people can use it at once. Members use the models free for the hours of their own plan (Base 50 a month, Pro 120, Max unlimited), then pay in credits, half going back to the members; traffic is always paid in credits. The group also lends to people outside it, its members first (on Pro and Max its maker can keep it private): the members share half of what it earns. The group's computers process the members' questions, so use a group only with people you trust.

BitPull

Build with what we have. Together.

Start with your computers. Connect your AI. Make room for what comes next.

Download the app↗Try it free, no card needed↗

03Shared code

A shared consciousness for developers' AI

Your AI finds code already written and described by others and reads only the files it needs. That can save tokens; how much depends on the project.

What one developer's AI has written and learned, every other AI can find and reuse. Right now 168,381 code repositories are ready: 0 shared by BitPull users and 168,381 from GitHub, and the number grows by itself. Your AI searches them first, reads the short description and fetches only the files it needs, instead of writing everything from scratch: fewer tokens, less time, and code others already use.

MCPClaude CodeCodexCursor
search_repos("jwt auth middleware")auth-kit★ 1.2kexpress-jwt-guard★ 860fastapi-login★ 430express-jwt-guard/read_manifest ✓README.mdsrc/middleware.ts✓src/verify.ts✓src/index.tstest/ (14)docs/ (31)your-project/+ middleware.ts+ verify.tsfetch_files(2)2 / 48 files
  1. 01Your AI searches the shared code
  2. 02It reads the short description
  3. 03It fetches only the files it needs

Open-source code from GitHub, ready for your AI

Next to the code shared by BitPull users, your AI tools find selected public GitHub repositories with a permissive or GPL licence, kept up to date. The files come straight from GitHub at a fixed commit, with their authors and licence.

GitHub → BitPull → MCP
01↗
02↗
03↗
</> + </> + </>

Connect your AI tool

Claude Code, Codex, Cursor and others reach BitPull through MCP.

npx -y bitpull mcp

llms.txt ↗ · install.json ↗

Frequently asked questions

Does BitPull include public code from GitHub?

Yes, from version 0.1.29. BitPull offers selected public GitHub repositories with a permissive licence (MIT, Apache, BSD and others) or a GPL licence, and keeps them up to date. Your AI tool finds them with the same search as the code shared by BitPull users, marked as coming from GitHub. The files are downloaded straight from GitHub at a fixed commit by your own computer, which leaves out sensitive files and hides keys; respect each repository's licence and credit its authors.

How does BitPull help save AI tokens?

Writing code from scratch makes an AI produce, and often correct, many thousands of tokens. With BitPull it first searches the shared repositories (168,381 right now), reads each one's short description and fetches only the files it needs, which usually costs far fewer tokens. Simple tasks can also go to open models on the network, paid in credits instead of your own AI's tokens. How much you save depends on the task.

BitPull

Build with what we have. Together.

Start with your computers. Connect your AI. Make room for what comes next.

Download the app↗Try it free, no card needed↗

04Our vision

Great ideas shouldn’t wait for a bigger computer.

There is knowledge in every project and potential in computers sitting idle. We are building BitPull to connect them, and make more possible together.

</> + CPU + AI
01

Use what already exists.

A useful piece of code should be easy to find. A computer with capacity to spare can help someone else. Sharing makes those resources go further.

02

Keep the choice with people.

You choose what to share and which computers to trust. For private work, use only your own computers. Community jobs can be read by the computer running them.

03

Work with your tools.

BitPull fits into the way you develop: your editor, your terminal, your AI assistant. Open models and MCP let you connect more of your work.

About us

BitPull was born from a simple problem: AI tools run out of credits right when you are working, while many computers sit switched on doing nothing. The power developers need already exists. What was missing was a safe way to share it.

BitPull is built by Zefirox S.R.L., an Italian software company based in Milan, founded in 2022. We want AI that costs less, that leaves you in control of your computers and your data, and that says plainly what it does and what it does not do. That is why we also publish our limits.

Security

Frequently asked questions

Is bitpull.cloud the same as other products called BitPull?

No. This BitPull is the developer network at bitpull.cloud and is not affiliated with other products or companies with a similar name. The official app is downloaded only from bitpull.cloud (or github.com/SamuealPK/bitpull-releases), the npm package is “bitpull” and the MCP server is io.github.SamuealPK/bitpull.

BitPull

Build with what we have. Together.

Start with your computers. Connect your AI. Make room for what comes next.

Download the app↗Try it free, no card needed↗

Plans

Paid plans, from €4.50 a month

Models run on the community's idle computers, not in a data centre paid for by your subscription. Paid plans start at €4.50 a month against about €20 for a typical AI subscription, there is a plan with no monthly fee, and lending is always free. Your own computers cost nothing, and you earn credits when others use them.

Base

€4.50/month
  • 100 credits every month for open models (≈ €6.25)
  • Groups of 4 computers, models up to 32 GB
  • 10 GB of group traffic through our server
  • 1 group, 50 hours a month with your groups

Pro

€9.90/month
  • 250 credits every month for open models (≈ €15.63)
  • Groups of 9 computers, any model, high priority
  • 30 GB of group traffic through our server
  • 3 groups, 120 hours a month with your groups, private groups

Max

€19.90/month
  • 600 credits every month for open models (≈ €37.50)
  • Groups of 9 computers, any model, first in the queue
  • 60 GB of group traffic through our server
  • Unlimited groups and hours, private groups

No monthly fee

Pay as you go

€0
  • No monthly fee: buy credits only when you need them
  • Groups of 2 computers, models up to 8 GB
  • 5 GB of group traffic through our server, then 0.5 credits per GB
  • 1 group; using it is paid in credits, earned by lending

Prices include VAT. Recharge any amount from €5, at 16 credits per euro. Plan credits work on open models; paid models use the credits you buy. Bought and plan credits are for your use on BitPull; credits earned by lending can be withdrawn as euros.

What do credits buy?

€5 buys 80 credits: about 8.8 million tokens on a small open model, 3.4 million on a medium one, 0.8 million on a large one, or 266.6666666666667 hours of one processor core.

What is a group?

Your own computers running one larger model together. The plan sets how many computers can join.

What do the GB mean?

The size of the model's file, not your graphics card's memory. Together your computers need a little more memory than that.

Can several people share an account?

An account belongs to one person, who can connect several computers.

BitPull and AI subscriptions

Costs less, because the computers already exist

Models run on the community's idle computers, not in a data centre paid for by your subscription. Paid plans start at €4.50 a month against about €20 for a typical AI subscription, there is a plan with no monthly fee, and lending is always free. Your own computers cost nothing, and you earn credits when others use them.

BitPull and AI subscriptions

ServicePrice per monthWhat you get
Pay as you go€0 + what you useOpen models on community computers, paid per use. Your own computers are free.
BitPull Base, Pro, Max€4.50 to €19.90Monthly credits for open models, and groups of your computers.
ChatGPT Plusabout €23OpenAI's models
Claude Pro, Google AI Proabout €18–23Anthropic's or Google's models

Prices seen in November 2026: check each provider. Open models are not the same as these companies' top models, and community computers can be slower than a data centre.

BitPull and AI coding tools

BitPull does not replace these tools: it works with them through MCP. They keep their own subscription, and BitPull adds open models and computers for routine work.

ToolMonthly plans, in dollars
Claude CodeClaude Pro $20 · Max $100 / $200
CodexChatGPT Plus $20 · Pro $100 / $200
CursorPro $20 · Pro+ $60 · Ultra $200
GitHub CopilotPro $10 · Pro+ $39
WindsurfPro ~$20

Prices before tax, as reported by secondary sources in November 2026: check each provider. Each plan has its own usage limits, which change often. Open models are not these companies' top models.

For lenders: earn credits with your idle computer

Lending is off until you turn it on. You choose when and how much, you earn 80% of what each job costs, and jobs have no access to your files.

Estimate your credits

An estimate at today's rates. You earn only when someone really uses your computer, so it can be much lower, even zero. Prices are set to cover the electricity of a computer working for others, valued as credits spent on BitPull: lending is for sharing, not for earning money.

In plain words

MCP
The standard way AI tools such as Claude Code connect to other services.
Sandbox
An isolated space where a job runs, with no internet and no access to the computer's files.
GPU
The graphics card: it runs AI models much faster than the processor.
Open model
An AI model whose files are public, so anyone can run it.
Credit
What you spend and earn on BitPull. €1 buys 16.

Frequently asked questions

Why does BitPull cost less than other AI services?

Open models run on computers people already own, so nobody pays for a data centre. Paid plans start at €4.50 a month, there is a pay-as-you-go plan with no monthly fee, questions to your own computers are free, and lending earns you credits. Paid models such as Claude or GPT cost their official price plus a margin: there the advantage is one balance for every model.

How much does BitPull cost compared with Runpod and GPU clouds?

Less for the same work, even when you pay with a recharge (1 credit ≈ €0.0625). An hour of an RTX A5000-class card with 9 cores and 25 GB costs 2.15 credits (≈ €0.13) against $0.16 on Runpod; an RTX 4090 3.95 credits (≈ €0.25) against $0.34; data-centre cards about 10% less than Runpod (an H100 about 24.6 credits, ≈ €1.54, against $1.99); a million tokens of a medium open model 23 credits (≈ €1.44) against $10 for Qwen3 32B. Only the RTX 3090, which draws much more power, costs about as much as on Runpod. Runpod rents data-centre machines with guaranteed availability; BitPull uses computers people lend, which may be busy, and the computer that runs a job can read it.

How are BitPull's prices calculated?

From the costs of the people who lend and from the prices of GPU clouds. For gaming graphics cards and open AI models the price covers the electricity of the computer doing the work, plus about 15%. We count Italian household electricity (€0.314 per kWh, Eurostat, first half of 2026; the EU average is €0.283), the power a whole computer draws with that card at full load (for example 550 W with an RTX 4090), the 80% of the price that goes to the lender and the 20% to BitPull, and credits valued as credits spent on BitPull (€0.0625). Data-centre cards, processors, storage, images and speech cost about 10% less than Runpod's Community Cloud for the same thing (dollars at €0.86). A GPU's price includes up to 9 cores and 50 GB of memory, as on Runpod. All these numbers are public at bitpull.cloud/public/plans (priceBasis) and change when electricity or Runpod's prices change.

Can I earn money by lending my computer?

A little, and that is not the point. Prices are set so the credits you earn cover the electricity your computer uses while it works for others, counted at Italian household prices (among the dearest in Europe), with a little left over. Withdrawn as euros a credit is worth less (0.02 € today), so with a gaming graphics card the euros usually do not cover the electricity; with the processor only, or a data-centre card, something is left. The real advantage is sharing: the computer you already have and the electricity you already pay become credits you spend on other people's computers (larger models, more power when yours is busy) instead of paying a subscription to a big company. Example: an RTX 4090 working for others for an hour uses about 0.55 kWh (≈ €0.17) and earns 3.16 credits, worth about €0.20 spent on BitPull.

How much does it cost to use my own group of computers?

Each member uses their groups free for the hours their plan includes: Base 50 hours a month, Pro 120, Max unlimited; on Pay as you go none are included. An hour counts only when it starts with one of your questions. Beyond that you pay in credits like anyone else, and half goes back to the group's members, you included for the part your computers hold. Pay as you go and Base allow 1 group, Pro 3, Max any number; Pro and Max can keep a group private.

BitPull

Build with what we have. Together.

Start with your computers. Connect your AI. Make room for what comes next.

Download the app↗Try it free, no card needed↗

05Earn

Your computer can earn. Even while you sleep.

When your computer is idle, people around the world use it for AI and code. Every job earns you credits: spend them on any model, or withdraw the ones you earned as euros to your bank.

while you sleep02:1403:4105:02Your computerzzzBitPulllending · idleTokyochat · 7B modelBerlinnpm test · sandboxSão Pauloimage · 1024 pxYour balance12.012.813.614.4+0.8+0.8+0.8Someone uses a modelpays 10 creditsqwen · 32BThe computer that workedBitPullservers, safety, new featuresthey can spend them, or withdraw themEarned14 days to matureWithdrawStripe checks and sends+ €Your banktransfer received€Your groupMac miniPCLaptop+0+1.6+0+1.6+0+1.6one big model, many computershalf to the membersSomeone outside the groupchat · 70BBitPull

At night your computer works for others: a chat in Tokyo, tests in Berlin, an image in São Paulo.

Whoever uses a model pays in credits, and most of them go to the computer that did the work.

Credits earned by lending mature, then Stripe sends them to your bank as euros.

A group of computers runs a model none could run alone, lends it to others and its members share half of what it earns.

Four ways to grow your credits

01 — 04
01

Lend your computer

You choose when and how much. Every job your computer does earns you credits, and jobs have no access to your files.

02

Lend with a group

A shared group lends its model to others too, members first. The members share 50% of what it earns, by memory and hours offered.

03

Share your code

Publish a repository: each new person whose AI fetches it gives you 0.1 credits, up to 50 a month.

04

A welcome gift

10 credits when you sign up, to try the models straight away.

03

Your piggy bank

Pick a computer, or set the processor, memory and graphics card yourself, and how many hours it is free. See what it could bring you in a month.

Or set the resources yourself:

Per day
Per month

An estimate at today's rates. You earn only when someone really uses your computer, so it can be much lower, even zero.

04

Where your credits go

One balance for everything, on your computer and on your phone.

  • Chat with open models
  • Claude, GPT and paid models
  • Images
  • Music and voice
  • Tests on many computers
  • Group traffic

05

The rules, plainly

Withdrawals in euros are not active yet. The rules are already in the Provider Terms, and what you earn stays in your balance.

Withdraw your earnings↗

Can become euros

  • Credits earned by lending your computer
  • Only from jobs completed and not contested
  • After 14 days of maturation

Stay credits

  • Credits you buy or get with a plan
  • Welcome, gift and shared-code credits
  • Earned credits you have already spent

To be paid you must be an adult with your tax residence where Stripe can pay out (European Economic Area, Switzerland, United Kingdom, United States, Canada). Stripe checks your identity and bank account, large amounts are checked by a person, and you declare what you receive.

Read the Provider Terms↗

Frequently asked questions

Can I earn money with BitPull?

Yes, by lending your computer when you are not using it: every job it does earns you credits. Credits earned by lending can become euros sent to your bank by Stripe, 14 days after the job and if nobody contested it. Bought and gift credits stay credits. What you earn depends on how much your computer is really used, and can be zero.

How do I withdraw what I earned?

On your account page choose "I want to be paid", accept the Provider Terms and open a Stripe account in your name, which checks your identity and bank account. When your earned credits have matured and pass the minimum shown, press Withdraw: Stripe sends the euros to your bank. Large amounts are checked by a person first. If a payment fails, the credits come back to your balance.

Which credits can become euros?

Only credits earned by lending your computer, from jobs that were completed and not contested, after the maturation period. Credits you buy, plan credits, welcome credits, gift credits and the credits for shared code are never paid out. Earned credits you spend can no longer be withdrawn.

How much can my computer earn?

It depends on its power, the hours it is free and how much people use it. The piggy bank on the Earn page gives an estimate at today's rates: for example a laptop with 4 cores and 8 GB, free 8 hours a day and used a quarter of that time, makes about 71 credits a month. It can be much less, even zero, because you earn only when someone really uses it.

Can I earn money by lending my computer?

A little, and that is not the point. Prices are set so the credits you earn cover the electricity your computer uses while it works for others, counted at Italian household prices (among the dearest in Europe), with a little left over. Withdrawn as euros a credit is worth less (0.02 € today), so with a gaming graphics card the euros usually do not cover the electricity; with the processor only, or a data-centre card, something is left. The real advantage is sharing: the computer you already have and the electricity you already pay become credits you spend on other people's computers (larger models, more power when yours is busy) instead of paying a subscription to a big company. Example: an RTX 4090 working for others for an hour uses about 0.55 kWh (≈ €0.17) and earns 3.16 credits, worth about €0.20 spent on BitPull.

How are BitPull's prices calculated?

From the costs of the people who lend and from the prices of GPU clouds. For gaming graphics cards and open AI models the price covers the electricity of the computer doing the work, plus about 15%. We count Italian household electricity (€0.314 per kWh, Eurostat, first half of 2026; the EU average is €0.283), the power a whole computer draws with that card at full load (for example 550 W with an RTX 4090), the 80% of the price that goes to the lender and the 20% to BitPull, and credits valued as credits spent on BitPull (€0.0625). Data-centre cards, processors, storage, images and speech cost about 10% less than Runpod's Community Cloud for the same thing (dollars at €0.86). A GPU's price includes up to 9 cores and 50 GB of memory, as on Runpod. All these numbers are public at bitpull.cloud/public/plans (priceBasis) and change when electricity or Runpod's prices change.

Is there a bonus for the first people who lend?

Yes: the first 100 people who lend get 50 credits once, after 10 hours of lending and one job done for someone else. They work on open models and are not paid out. The app's panel shows the hours so far and the places left; /public/plans shows the places left.

Turn your idle computer into credits

Download the app, sign in and turn on lending in the panel. You can pause it whenever you like.

Download the app↗See plans and prices

or

Your account

Payment received. Credits and plan update within a few seconds.
Credit balance
–
Plan
–
Connected computers
–

Buy

Install the app

The app runs on your computer: it shares the folders you authorise and lends resources by your rules.

Download

Your published repos

NameDescriptionLicense

Credit history

DateReasonCredits

Get paid for lending

Credits you earn by lending your computer can become euros, sent to your bank by Stripe. Bought and gift credits stay credits.

Bought and gift credits
–
Earned by lending
–
Redeemable now
–
Maturing
–
Being paid
–

Your payouts

DateCreditsEurosStatus

Download BitPull

The app connects this computer to the network, opens your dashboard and plugs into Claude Code, Cursor and other AI tools.

The app is not distributed through app stores. On macOS, the first time: open the app, then go to System Settings → Privacy & Security and click Open Anyway next to BitPull. On Windows, choose More info → Run anyway.

To lend computing power you also need Docker (or Apple Container on macOS); to lend AI models, Ollama.

Prefer the terminal? The same agent runs as a command-line tool.

Install from the terminal or with your AI

The script downloads the latest version, checks its SHA-256 and installs it for your user. macOS and Linux first, Windows (PowerShell) second.

curl -fsSL https://bitpull.cloud/install.sh | sh
powershell -ExecutionPolicy Bypass -c "irm https://bitpull.cloud/install.ps1 | iex"

AI assistants (Claude Code, Codex, Cursor…) find everything they need to install and set up BitPull here: llms.txt · install.json

Troubleshooting

Common problems with BitPull and how to solve them, from the exact message you see to the fix. Kept up to date with every version.

To check a specific computer run bitpull doctor · help.json · llms.txt

Connect a computer

Your BitPull app, command line or AI assistant asked to log in to your account. Enter the code it shows.

Approve only if you started this yourself on your computer just now. Never approve a code someone sent you.

Report content

Report illegal content, malware, copyright infringement or abuse. We review every report and reply if you leave an email.

How BitPull protects you

What is protected, what is not, and what you can do about it, limits included.

Private data? Use only your own computers. A computer that runs a job can read what is in it. For anything private choose "Only my computers": it never reaches anyone else, and it is free.
After a job, your data is deleted Your data travels encrypted and is opened only inside an isolated space on the computer that runs the job. BitPull gives the person lending no tool to see it. As soon as the result is delivered, it is deleted from that computer, which does not keep the key to read it again; the encrypted copies in transit are deleted for good within 24 hours at most. If you lend: the job cannot read your files or use the internet, and leaves nothing on your computer. If the app or the computer stops in the middle of a job, what it left is removed before the next one starts.

Who sees what

You
Encrypted
BitPull server
Encrypted
Computer running the job
WhoCan seeCannot see
BitPull serverSize, model, time and cost of a jobWhat is inside the job
Computer running the jobWhat is inside that job, while it worksYour account, your other jobs, your files
Paid models (Claude, GPT…)Your request, which passes through our server to the providerYour jobs on the network

What is encrypted

  • Job content is encrypted on your computer with a key made for that job. Our server keeps only the encrypted copy, and the key is sealed so that only the computer running the job can open it.
  • Results come back encrypted the same way, and the encrypted data is deleted 24 hours after the job ends.
  • While it works, the computer running the job has to read the content. Encryption cannot change that.
  • Requests to paid models pass through our server to the provider and are not protected this way.

If a computer misbehaves

  • It can read the job it receives and can return a wrong answer.
  • It cannot reach your account, your computer or other people's jobs.
  • It never receives jobs kept for your own computers: our server decides this, not the app.
  • Computers that fail or give wrong results lose reputation, and below a limit they stop receiving jobs.
  • Some jobs with a fixed result are run twice and compared. AI answers cannot be checked this way.

How lenders are protected

  • Jobs run in a sandbox with no internet, no access to the computer's files and no system privileges, with limits on processor and memory. It is deleted when the job ends.
  • Lending is off until you turn it on, and you choose when, how much and what to lend.
  • Open AI models run in their own program, not in a sandbox: they receive text and files to read, not commands to run.

Groups of your computers

  • Your computers talk over an encrypted connection with a key only they have, directly on a local network or through a relay that cannot read it.
  • While a group model runs, it is shared with the community, unless the person who made the group keeps it private (Pro and Max plans); its owner and members have priority. Group inference runs outside the job sandbox. Use groups only with people you trust. A custom Hugging Face model stays private.

The limits

  • No sandbox is perfect: it lowers the risk, it does not remove it.
  • Our server hands out the keys that protect jobs, so you are trusting us to run it honestly.
  • Anyone with your login can act as you: use a strong password.
  • BitPull has not had an independent security audit yet.

This is a translation provided for convenience. The English version is the binding one.

BitPull.cloud
ProductNetworkShared codePlansEarnOur visionTerms of ServicePrivacy PolicyAcceptable Use PolicyCredits and BillingLender TermsProvider TermsReporting and Notice-and-ActionCookies and Local StorageReport contentHelpSecurityFor AI assistantsWhat's newService statusComparisonsWho BitPull is for
Zefirox S.R.L. · Via Giambellino 79, 20146 Milano (MI) · P.IVA 12570720966 · REA MI-2669986 · PEC zefirox.srl@pec.beapec.it