LLM Guide
The LLM guide is one plain-text file, written for AI assistants, that teaches the OnlyWorlds standard, the 22 element categories, the API and the TypeScript SDK. Paste it into a chat and the assistant can help you plan and build your own tool on OnlyWorlds.
The file: https://onlyworlds.github.io/assets/ow_llm_guide.txt
The guide is for people building their own tools. Someone who wants to build a world can go straight to Atlas, and someone working in Claude Code is better served by the toolkit, which the guide itself points to.
Using It in a Chat
Section titled “Using It in a Chat”- Open the file and copy all of it, or download it.
- Paste it into a new chat (ChatGPT, Claude, or a browser builder such as Lovable, Replit, v0.dev or ChatGPT Canvas).
- Describe the tool you want. The guide tells the assistant to ask what problem it solves, who uses it, its one core interaction, and which element categories matter, then to write a short spec with you before writing code.
For full field definitions beyond the guide, point the assistant at the schema or at FIELD_SCHEMA in the SDK.
What It Covers
Section titled “What It Covers”| Part | Contents |
|---|---|
| 1. How to help the user | Gauging experience (no coding, some coding, advanced), planning a tool before building it with a short spec, choosing between browser builders and a local setup, pointing Claude Code users to the toolkit, and local setup (Node.js, Git, an editor). |
| 2. OnlyWorlds essentials | What OnlyWorlds is (an open standard; a world as a folder of JSON files or a hosted world with a REST API), the 22 categories in six groups with one line each, how common concepts map onto them, and x_ extension fields for custom data. |
| 3. Credentials and first call | Minting a key (ow_w_, ow_r_, ow_a_, legacy), where the PIN is set and when it is sent, the demo keys, keeping credentials out of chats, commits and browser bundles, and a first SDK call. |
| 4. SDK and API reference | Plain HTTP (the base URL, the API-Key and API-Pin headers, the error envelope); reads and cursor paging, filters, expand and fields; create, patch, upsert and delete; link fields and editLinks; bulk upload with client-minted ids and idempotency keys; the change feed; error codes; UI helpers (icons, colours, field groups, FIELD_SCHEMA); reading a world folder with no account. |
| 5. Deployment | Building, the static hosts the API accepts browser calls from without setup, deploying to Cloudflare Pages, and why a public site uses a read key. |
| 6. Resources | Links to Atlas, the developer start, the API reference, the SDK, the standard, schema-dist, the toolkit, the MCP server, the tools and the community. |
The file ends with its own changelog.
Related
Section titled “Related”- MCP server: connect an assistant to a world directly instead of pasting the guide.
- Toolkit: the same knowledge as Claude Code skills, readable by other agents as markdown.
- AI agents: the four ways an AI works with a world.
