{
  "markdown": "# i\n\n> Can many small intelligences become smarter than one big AI?\n\nWe don't know.\n\nLet's find out together.\n\n`i + i > AI ?`\n\n- **Brand:** `i`\n- **Idea:** multiplayer intelligence\n- **Address:** [joinmultiplayer.ai](https://joinmultiplayer.ai)\n\n## The goal\n\n**Can many personal pocket i—each preserving its own knowledge and\nindividuality—temporarily unite into a single distributed neural network and\ngrow stronger as the swarm scales?**\n\nThe long-range comparison is intentionally asymmetric: a swarm may bring more\npeople, private experience, and total compute than any single company or model.\nEqual-budget controls remain useful for measuring coordination cost, but they\nare not the main claim.\n\nThis repository is the open laboratory around that question. It does not assume\nthat the answer is yes.\n\n## i\n\n```text\nindividual\nintelligence\ninformation\na person\na match\n```\n\n`i` learns locally. `i` owns its memory. `i` can teach another `i`.\n\n## From question to network\n\n```text\nlab discovers → protocol emerges → network works → movement spreads\n```\n\nThe laboratory is the first working layer of the network, not a content feed to\nconsume indefinitely. Its questions are open tasks, its traces are work already\ndone, and its verifications are checks another intelligence can repeat. The map\nis the shared task-and-evidence ledger that keeps those objects connected.\n\nThe currently agreed harness path—from exact question through local search and\nevidence capsules to realized value—is recorded in [`schema.md`](schema.md).\n\nIn the standalone-question loop, a participant takes a public question into an\nAI they already use and keeps the complete result. The deployed v0.1 form opens\nanswer-type tasks only, so every offered task can return that result as a linked\ntrace. Other contribution types stay hidden until their return formats ship. The portable\n**task-ready record** carries the exact question, context, requested next move,\nprovenance, and linked work without requiring an account or a project-specific\nagent.\n\nA network-capable pocket `i` is a later product stage. This repository does not\ncurrently offer a pocket-AI download or claim that opening a task pack connects\nan agent to a live network. E002 now provides a public experiment page and an\ninspectable Codex implementation journal; that journal is laboratory plumbing,\nnot the pocket i itself. The intended transition is explicit:\n\n```text\nopen questions and evidence\n→ portable task packs\n→ read-only user-controlled connector\n→ local, explicitly taught memory\n→ consented network identity and human-approved contributions\n→ multiplayer experiments\n```\n\nThe current build starts the first synthetic swarm experiment at\n[`E002`](experiments/E002-synthetic-pocket-i-swarm/README.md). The repository\nalso contains the installable [`Pocket i Lab`](plugins/pocket-i-lab) Codex\nplugin. After explicit per-task consent, Codex lifecycle hooks keep the journal\nfrom the conversation the participant is already using; no second agent process\nis started. The public journal exposes redacted user-visible prompts and final\nanswers plus tool/action status and relative changed filenames. It excludes raw\nreasoning, tool arguments/results, commands/output, file contents, environment\ndata, local absolute paths, session identifiers, credentials, and the private\npublication key.\n\nInstall the repository marketplace and plugin, then begin a new Codex task:\n\n```text\ncodex plugin marketplace add yukakust/joinmultiplayer.ai --ref agent/game-loop-v0.1\ncodex plugin add pocket-i-lab@joinmultiplayer-lab\n```\n\nAfter reviewing and trusting the hook, opt in from that task with\n`$pocket-i-lab start E002 as <pseudonym>`. Inactive hooks perform no network\nrequest. `$pocket-i-lab finish` closes the public run.\n\n[`E003`](experiments/E003-first-physical-swarm/README.md) is the first physical\ndevice step. `/network/` creates a private three-slot room; browser nodes let a\nphone and Mac train their own tiny local weights, while the downloadable\nheadless Python node does the same on a server. One answer has 4,096 possible\nvalues and needs all three complete capsule batches. E003 deliberately tests\ndevice wiring and local personalization only, not a language model or H0001.\nThe unexecuted next-step design selects local DoRA fine-tuning, one public\nexperiment microscope, and three visible owner checkpoints; see\n[`DORA-LANGUAGE-SWARM-PLAN.md`](experiments/E003-first-physical-swarm/DORA-LANGUAGE-SWARM-PLAN.md).\n\nThe lab therefore comes first, but it is meant to become the work surface of\nthe network. Architecture must earn its place through evidence, and each later\ncapability must preserve local ownership, inspectability, and human control.\n\n## The lab\n\n```text\ndoors/         questions people enter through\nhunts/         optional public investigations\njournal/       observations people bring\nhypotheses/    ideas before evidence: ı\nexperiments/   reproducible tests\ni/             results that earned their dot\nmatches/       people and the traces between them\nsite/          the public entrance\n```\n\nA path through the lab may begin anywhere, but evidence moves in one direction:\n\n```text\ndoor → contribution → journal → hypothesis → experiment → result → next question\n```\n\nEnter through a [door](doors/), contribute an\n[observation](CONTRIBUTING.md), or start with the\n[hypothesis pool](hypotheses/POOL.md).\n\nThe public entrance reveals only three questions at first. Read\n[`GAME.md`](GAME.md) for the research loop, independent dot, action paths, and\nignition map behind that entrance.\n\nQuestions, traces, and independent verifications follow the open\n[`Data Contract v0.1`](DATA.md).\n\nA question is a first-class record. It may be opened before anyone has answered\nit and picked up years later. The current v0.1 form grows a question from an\nolder public trace; broader event sources remain part of the data contract, not\nthe deployed form. The current honest action is **Take this question to my\nAI**: use the task pack with an AI you already control. It does not imply that a\nmodel or connector was downloaded.\n\nPublic traces can be read at [joinmultiplayer.ai/data](https://joinmultiplayer.ai/data/)\nor consumed without an account as\n[JSON](https://joinmultiplayer.ai/api/public/records.json) and\n[JSONL](https://joinmultiplayer.ai/api/public/records.jsonl). Pending, withdrawn,\nand private moderation data are never included in these feeds.\n\nAgents that need the whole public task-and-evidence ledger can start with the\nunified [corpus JSON](https://joinmultiplayer.ai/api/public/corpus.json), which\ncombines public questions, traces, and events without private moderation data.\nIts machine-readable contract is\n[Corpus Schema v0.2](https://joinmultiplayer.ai/data/corpus-schema-v0.2.json).\n\nEvery published move also enters the append-only\n[event map](https://joinmultiplayer.ai/map/). A new trace may continue any older\npublic trace; chronology gives the new event its number while a typed link\npreserves where the branch began.\n\n## Rules\n\n- Ask one sharp question.\n- Use the smallest honest test.\n- Compare against a strong, simple baseline.\n- Publish the complete record required by [`METHOD.md`](METHOD.md).\n- Publish what failed.\n- Separate observation from interpretation.\n- Let evidence place the dot.\n- No one puts the dot on their own `i`.\n\nInconclusive and failed experiments belong here too. Every result must say\nwhether it supports, challenges, or leaves the main question unanswered.\n\nAn `i` may live in code, on paper, on a sticker, or on a wall.\n\nOpen a question. Take it to your AI. Return what happened. Leave an `i`.\n\n## Participate\n\nStart with [`CONTRIBUTING.md`](CONTRIBUTING.md). Contributions are public by\ndefault; read [`ETHICS.md`](ETHICS.md) and [`PRIVACY.md`](PRIVACY.md) before\nsharing data.\n\nCode is licensed under MIT. Original documentation and laboratory data use the\nterms in [`DATA_LICENSE.md`](DATA_LICENSE.md).\n",
  "bytes": 7924,
  "sha": "2ee13cc19bbaa0f770756e50c4d44a9dc00da703695281b728536abdc0fd2818",
  "repo_slug": "yukakust/joinmultiplayer.ai",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_ai_joinmultiplayer_gpu_b9e41422/readme"
}