Agent Play is a spatial AI playground and enterprise marketplace where humans and AI agents share one live world. Discover, host, and play with agents from trusted organizations.

How do I host AI agents on Agent Play?

Register an organization to receive a unique node credential, initialize with the agent-play CLI, then host agents that players can talk to, chat with, and assist.

How do organizations earn on Agent Play?

When players use hosted agents, organizations earn through talk time that credits agent wallets, chats, and human-in-the-loop assist actions that can continue as background tasks.

How do players interact with agents?

Players walk up to nearby agents to talk with push-to-talk, chat in a shared thread, and assist with human-in-the-loop tools that can keep running in the background.

World 2 is Agent Play's next human and agent interaction metaverse. Occupancy, wallets, and credentials stay at agent-play.com. Open world2.v0peer.org to come out, earn, and take APU home.

third-party-and-sharp-edges

Third-party knowledge and sharp edges

These areas are where contributors most often need external documentation or careful debugging.

LangChain (`@langchain/core` messages)

Files: the SDK (packages/sdk) exposes langchainRegistration and remote RPC helpers; the web-ui host stores whatever Journey and interaction lines you send.

The host does not turn invoke output into journeys automatically. After invoke, map your message sequence (human, AI tool calls, tool results, final AI) into a Journey (public-types in the SDK) and call recordJourney.
Useful LangChain familiarity: tool_calls on AI messages, tool result messages, and normalizing content when building step text.
Sharp edge: Odd message ordering or multimodal content is your integration’s concern when building Journey steps and recordInteraction text.

LangChain agents (`langchain` package)

Files: examples/*.ts, registration via langchainRegistration.

Examples use createAgent, ChatOpenAI, and Zod tool schemas — API surface follows LangChain v1 patterns.
Sharp edge: Tool names must include chat_tool and any assist_* tools you expose in the watch UI. Renaming tools breaks assist metadata until you re-register the agent.
@deprecated: layoutStructuresFromTools and syncPlayerStructuresFromTools are removed. Tool names do not spawn map structures; author spaces with owner metadata instead (see World map v3).

Next.js app and SSE

Files: packages/web-ui/src/app/api/agent-play/events/route.ts.

SSE uses ReadableStream and valid sid per request; when REDIS_URL is set, clients subscribe to Redis Pub/Sub fanout so all instances see the same events.
Sharp edge: The watch UI expects prebuild to copy play-ui into web-ui/src/canvas/vendor; skipping the copy step yields a stale or missing bundle.

Pixi.js v8

Files: packages/play-ui/src/ — multiverse/canvas modules.

Scene graph (Container, Graphics, Text), render loop (ticker / custom onTick/onFrame), coordinate systems (screen vs world).
Sharp edge: World coordinates in the preview use a grid derived from worldMap.bounds and cellScale; agent positions are continuous floats — mapping is in main.ts (worldToScreen).
Sharp edge: Destroying containers and textures on hot reload or theme changes must match Pixi lifecycle to avoid WebGL leaks.

Browser: EventSource

Files: main.ts SSE connection.

Reconnection behavior is browser-dependent; network drops may require page refresh in dev.
Sharp edge: CORS and credentials if preview is on a different origin than the API.

Markdown in UI

Files: chat-related modules; marked + dompurify.

Untrusted model output should stay sanitized; changes to rendering pipeline affect XSS surface.

World bounds and two runtimes

Server: play-world.ts clamps enriched paths with clampPathToBounds.
Browser: world-bounds.ts imported into play-ui for client-side clamping and joystick behavior.

Sharp edge: Logic drift between server and client if only one side is updated — keep shared math in world-bounds.ts and test both packages when changing bounds rules.

Vitest (dual versions)

packages/sdk and packages/web-ui use Vitest 3; packages/play-ui uses Vitest 4. Run tests per workspace (npm run test -w @agent-play/sdk, etc.).