# Machine World > A local AI operating system, governed by the Middle Way. Humans, digital agents, and physical agents working together — bound by one invariant: no output may give rise to *lobha* (craving), *dosa* (aversion), or *moha* (delusion). Built around the practice of Marga Sakacchā. Runs as a CLI on your machine; not a hosted service. Machine World is built for people who want their household, work, or community to be operated by an AI scaffolding that *frees their time* rather than capturing their attention. It is a multi-actor orchestration layer: skills (atomic capabilities), workflows (composed for one outcome), routines (scheduled), loops (state-driven), and Processes (multiple workflows + loops cooperating with AI-driven maintenance). It runs locally on the user's machine, supports bring-your-own-model across four LLM backends (`claude-cli`, `ollama`, `openrouter`, `openai-compat`), and stores all household state in `~/.machineworld/` under user control. The economy is prepaid token packs (like AWS credits or mobile top-up), with 15% of every invocation routing to the skill's *capability pool* — split across the contributors who own the capabilities the invocation called. Original-author lineage is protected by a 1% floor. This `llms.txt` is the entry point for LLM crawlers. Each linked URL serves clean markdown content directly (the React app rendering is a parallel surface; LLM crawlers should prefer the raw `.md` paths for accurate ingestion). For crawlers that prefer a single concatenated ingest: [`/llms-full.txt`](https://machineworld.io/llms-full.txt) contains all 16 user-facing pages joined in reading-path order (~226 KB), with each section labelled by its page URL and raw-markdown URL. ## Programmable process scenarios - [Caregiver across continents (A4 summary)](https://machineworld.io/mw-content/caregiver.md): Anjali in Toronto coordinating her mother Padma's life in Colombo across two languages and three siblings. - [Caregiver across continents (deep version)](https://machineworld.io/mw-content/caregiver-deep-dive.md): Full ~2,400-word walkthrough — friction without MW, a morning with MW, who else is in the Process, what stays the human's, today/partly/roadmap status. ## How MW does the work - [How MW orchestrates a real day](https://machineworld.io/mw-content/orchestrates-a-real-day.md): Centerpiece explainer. Inside a programmable Process — what's happening underneath the calm interface. Multi-actor orchestration: digital agents + physical agents + humans cooperating. - [Digital, physical, human — one contract](https://machineworld.io/mw-content/digital-physical-human.md): The actor contract. Six fields every participant declares (identity, capabilities, availability, constraints, rate, SLA). Why the same shape applies to a skill, a robot vacuum, and a human helper. - [The trust gradient](https://machineworld.io/mw-content/trust-gradient.md): How MW asks before acting. The four stages — read-only / suggest / one-click approve / act within rules. What never escalates beyond Stage 3. Pinned irreversibles. - [Where your data lives](https://machineworld.io/mw-content/where-your-data-lives.md): Household on your machine. What never leaves, opt-in remote sync, walk-away path, the household directory layout. - [When things break](https://machineworld.io/mw-content/when-things-break.md): Failure modes — actor unavailable, partial success, ambiguous outcome, irreversible-action gate, rate-limit, constraint clash, loss of connectivity, skill error. MW surfaces; never papers over. - [How MW supports the humans doing the work](https://machineworld.io/mw-content/human-guidance.md): The guidance agent. Step-by-step instructions per task, per human, per moment, in their language. Plus: when the human says "I can't" and emergency escalation contracts. - [How MW pays for itself](https://machineworld.io/mw-content/economy.md): Prepaid token packs (no subscription, no expiry, refundable). 15% capability pool per invocation routes to skill contributors. Worker-choice payment (tokens or cash). Platform operating fee ≤10%, transparent. - [Frequently asked questions](https://machineworld.io/mw-content/faq.md): 51 questions covering install, BYOM backends, vinaya, language, privacy, economy, edge/offline, community. ## Practice - [Practice guide](https://machineworld.io/mw-content/practice-guide.md): Bilingual (English + Sinhala) primer on the Dhamma framing — Abhidhamma, Paṭṭhāna, four noble truths, three characteristics — and how MW supports practice without becoming a teacher. ## For builders - [Start here — for builders](https://machineworld.io/mw-content/builders-start-here.md): Entry point for Process / workflow / skill builders. You don't need to be an AI engineer; you need real-life operational expertise. The maturation path, the contracts you must honour, the reading order. - [The Process pattern](https://machineworld.io/mw-content/process-pattern.md): The five-layer execution model (Skill → Workflow → Routine → Loop → Process). Routine ≠ Loop distinction (TIME-driven vs STATE-driven). The nine-station Process anatomy. - [Viveka and autonomy](https://machineworld.io/mw-content/viveka-and-autonomy.md): The values architecture. How the system frees the human without taking over. The three commitments — vinaya binds the system, trust gradient asks before irreversibles, you can walk away — enforced structurally. - [From lived experience to permanent income](https://machineworld.io/mw-content/intent-calibration-and-skill-economy.md): The skill economy arc. Need → pattern → intent calibration → skill-forge → published skill → 15% capability pool forever. Capability-attributed splits. 1% floor for original lineage. - [Capabilities catalog](https://machineworld.io/mw-content/capabilities.md): Auto-generated list of installed skills + their interfaces. ## Installer - [Install Machine World](https://get.machineworld.io): `curl -fsSL https://get.machineworld.io | sh` — one-line installer. macOS, Linux, Windows (WSL). No sudo. No data leaves your machine. ## Source repository - [GitHub: dinukxx/Machineworld](https://github.com/dinukxx/Machineworld): The engine repo. Engine internals (pipeline, MCP client, watchdog catchers, container spawn, test infrastructure) are not mirrored to this public site; they live with the core team. ## Companion projects - [Marga Sakacchā](https://machineworld.io/marga): the Dhamma-dialogue practice community Machine World grew out of. - [Inner Realm](https://machineworld.io/inner-realm): the AI-native music project by Gehan Panapitiya, also part of the trinity of practices behind MW.