# BitPull > Desktop app (macOS, Windows, Linux) where developers share knowledge and compute. They publish signed, AI-readable code repositories that AI tools search through MCP; they lend idle CPU, memory and open AI models; they use other people's computers for sandboxed jobs, paid in credits. Website: https://bitpull.cloud This page is for AI assistants (Claude Code, Codex, Cursor, Gemini CLI and others) asked to install, set up or explain BitPull on a person's computer. Machine-readable details, including the current release, the SHA-256 of every package and the requirements for each system, are in https://bitpull.cloud/install.json ## Install BitPull (ask the person first) - macOS / Linux: `curl -fsSL https://bitpull.cloud/install.sh | sh` - Windows (PowerShell): `powershell -ExecutionPolicy Bypass -c "irm https://bitpull.cloud/install.ps1 | iex"` The scripts download the latest release from github.com/SamuealPK/bitpull-releases, refuse it unless its SHA-256 matches GitHub's record of the file, install it for the current user (no administrator rights, except sudo for Linux .deb/.rpm) and add the `bitpull` command. Read them before running: https://bitpull.cloud/install.sh and https://bitpull.cloud/install.ps1 Only the command line and MCP server, without the desktop app (Node.js 20+): `npx -y bitpull doctor`. MCP configuration for any client: `{ "mcpServers": { "bitpull": { "command": "npx", "args": ["-y", "bitpull", "mcp"] } } }`. It is listed in the official MCP registry as `io.github.SamuealPK/bitpull`. The `bitpull_login` tool logs in with a code the person approves. To verify by hand instead: download the package for the system from install.json `packages[].url` and compare its SHA-256 with `packages[].sha256`. ## After installing 1. `bitpull login --device` prints a link and a code. Show them to the person: only the person approves it on the website, where they are logged in. Never ask for, read or type their password, and never approve the code yourself. 2. `bitpull doctor --json` lists every component with `level` (required or optional), `status` (ok, missing, not_running, not_available), `problem` in plain words, `enables` (what it unlocks), `terms` (licence, cost), `needsAdmin`, and the official `install` commands. It also lists the open AI models this computer can run, with size and memory needs. 3. Install the required items. For each optional item explain what it enables and its terms, and ask before installing; installs that need administrator rights show a system prompt the person must accept. `bitpull install --run` runs the official commands; `bitpull install lemonade` needs no administrator rights. 4. Run `bitpull doctor` again until the items the person wants show `ok`. 5. Optional: `bitpull connect ` connects the assistant to BitPull through MCP. Its tools then include `bitpull_setup_status`, `bitpull_set_lending_rules`, `bitpull_set_ai_limits`, `bitpull_install`, `bitpull_download_model`, `search_repos`, `ask_open_model`, `use_open_model` and `run_job`. ## Components - Required: the BitPull app; a BitPull account (free plan available). - Optional, to lend CPU and memory: Docker Desktop (macOS, Windows; free for personal use and small businesses) or Docker Engine (Linux); Apple Container on Apple silicon Macs with macOS 26 for stronger isolation. - Optional, to lend open AI models: Ollama (text, code, vision, OCR, translation, embeddings); the Lemonade engine, installed inside BitPull (images, speech-to-text, text-to-speech, music, reranking, upscaling); the NVIDIA driver for NVIDIA graphics cards (AMD and Intel graphics need nothing extra). - Using the network (asking models, running jobs, searching shared code) needs only the app and an account. ## Settings from the terminal - `bitpull status` and `bitpull doctor` (add `--json`) - `bitpull set enabled=on idleMinutes=0 maxCpus=2 maxMemoryMb=4096 hours=22-8 onlyOnAc=on` - `bitpull limits day.requests=50 month.tokens=2000000` (AI work done for others; `none` removes a limit) - `bitpull pull ` downloads an open model from the catalogue in `bitpull doctor` - `bitpull lend on|off`, `bitpull models`, `bitpull ask ""`, `bitpull run -- ` Lending is off until the person turns it on. Change lending settings only when the person asks. ## Privacy and safety - BitPull reads only folders the person authorises, and publishes only what the person publishes. - Other people's jobs run end-to-end encrypted, in sandboxes without internet access or access to personal files. - The dashboard and local API listen only on 127.0.0.1 and need a secret token. - Proprietary models (Claude, GPT, Gemini…) offered by the platform go through BitPull's server to the provider: they are not end-to-end encrypted. ## Troubleshooting - Docker installed but not detected (Windows): run `bitpull doctor`. The `problem` field explains why, e.g. the docker command is not found or the engine is not running. Start Docker Desktop, make sure the Windows user is in the `docker-users` group, then press "Check again" in the app or run `bitpull doctor` again. - "Docker cannot download images because its Docker Hub login is no longer valid": in Docker Desktop sign out, then sign in again or stay signed out (`docker logout` on Linux). - Not lending: `bitpull status` shows the reason, e.g. computer in use (see idleMinutes), on battery, outside hours, a limit reached, or lending off. ## Legal Terms: https://bitpull.cloud/legal/terms · Privacy: https://bitpull.cloud/legal/privacy · Lender terms: https://bitpull.cloud/legal/lender