An open standard defining how games stay alive, remember, and connect with people, reality, and AI.
No. Loom is a meaning layer that sits on top of engines. It doesn't render, doesn't compute physics, doesn't generate code. It defines the semantics of how games adapt, persist, and integrate with players and reality.
No, it's a protocol — a set of file formats and integration patterns. Like HTML isn't a website; it's the format websites use.
A weaving loom takes threads and creates fabric. Loom Protocol takes signals (real-world threads) and weaves them into game experiences. Threads + Loom + Fabric = Signals + Loom + Game.
The standard is free (CC BY 4.0). Implementing it has costs: AI Director LLM calls, signal API subscriptions if any, hosting. Loom doesn't impose its own fees.
Those generate worlds (visual, navigable spaces). Loom describes how those worlds stay alive, remember, and adapt. Loom complements them — you can use Genie for world generation and Loom for the meta-layer.
Inworld is one possible implementer of Loom's "AI Director" concept. Inworld creates LLM-driven NPCs; Loom standardizes the interface by which any such system reads game/player state. They can coexist.
| Steam | NFTs | Loom | |
|---|---|---|---|
| What it stores | Stats | Asset ownership | Identity + meaning + chronicle |
| User control | Valve | User (token) | User (DID) |
| Cross-platform | No | Sort of | Yes (open standard) |
| Cross-game semantic continuity | No | No | Yes |
Yes. Loom is engine-agnostic. You add an LGD file, integrate an AI Director (via MCP or your own LLM), and start using REALITY signals. Your engine handles rendering, physics, and gameplay code as before.
- Valid LGD with
IDENTITY,SOUL,WORLD,RULESlayers - That's it. You don't need PLGD support, REALITY signals, or AI Director for Bronze.
No. Local-first hosting is the default. Blockchain is optional and used only for DID anchoring and Chronicle moment hashes — never for body content.
YAML is canonical. JSON, TOML are accepted. Natural-language fields can be in any human language; spec uses BCP 47 tags.
Yes. MCP is model-agnostic. Director can use Claude, GPT, Gemini, self-hosted Llama, etc. Choose per task.
8 layers: identity (DID), soul (essence), patterns (behaviors), preferences (declared), contexts (current), boundaries (safety), chronicle (memories), permissions (control).
By default, on your device (local-first). Optional encrypted backups to IPFS, DWN, or trusted hosts. You always hold the keys.
Only with explicit permission, and only fields the game declares it needs. Loom's data minimization principle prohibits over-collection.
Yes. loom plgd erase --did ... triggers full local deletion, requests cloud backup deletion, and revokes blockchain anchors. (Note: blockchain data can be marked revoked but not physically deleted.)
Two options: (a) the game must work without that data (Loom requires fallbacks), or (b) you don't play that game. Loom prohibits coercive data demands.
No. PLGD is mutable. SOUL can change (with chronicle record). PATTERNS can be reset. People aren't fixed; the standard reflects this.
Same as any technology — yes, theoretically. But:
- The standard explicitly prohibits payment-pressure adaptations
- SOUL inviolability prevents "engagement-optimized" manipulation
- Conformance audits (Gold/Platinum) check for dark patterns
- Open spec means abuses are visible
We can't prevent all misuse, but the standard tries hard to discourage it.
These would fail Gold conformance audit. They contradict Loom's anti-dark-pattern principles. Bronze conformance might let them slip; Gold/Platinum requires ethics audit.
That's a legal question per jurisdiction. Loom encourages local-first hosting + encryption to make this difficult. End-to-end encrypted backup means even hosts can't comply with demands.
PLGD.identity.age_tier: child triggers strong protections: Tier 3-4 signals blocked, no gore/violence, special inheritance rules for parents. Aligns with COPPA, EU DSA.
As of v0.1 Draft (2026): no one in production. This is a proposal to define the standard early. Reference implementations and early adopters expected v0.5+.
- Indie devs: independence from closed platforms
- AI engineers: standard input for AI Directors
- Players: identity that travels
- Regulators: ethical framework already baked in
- Researchers: reference architecture
Adoption is a chicken-and-egg problem. Standards take time. HTML wasn't built in a day.
- v0.1 (now): Draft spec
- v0.2: Binding DSL grammar, validator
- v0.3: Translation Layer refinement
- v0.4: DID/VC, registry beta
- v0.5: Reference implementations
- v0.7: Foundation established
- v1.0: Stable RFC
Roughly 2 years if we move steadily.
Yes! See CONTRIBUTING.md. We need: spec critiques, translations, signal specs, validator implementations, ethics review, real-world testing.
Yes. The same protocol fits digital twins, training simulators, educational software, therapeutic games. See scenarios.md.
Partially. The PLGD concept (portable user identity) is generalizable. The LGD concept assumes a "world with rules" so doesn't directly fit pure conversational AI. But there's overlap.
The PLGD model could inform self-sovereign social identity. But Loom is scoped to game-like experiences. Adjacent standards (Solid, ActivityPub) handle social media better.
Human readability. LGDs/PLGDs are read by people (designers, players, auditors), not just machines. YAML's comments, clean syntax, and multi-line strings serve this. JSON is also accepted.
The author is Korean and the project starts in a Korean context. The standard is fully internationalized; Korean is just the reference language alongside English. Translations to other languages welcome.
Loom doesn't take a position on AI art ethics. It's about how games are structured, not how they're created visually. AI Directors can be highly involved or barely used per game.
Yes — the spec is CC BY 4.0, you can build commercial products on it. The Loom name and logo are reserved (use a different name for forks).