World instructions

Every world has world instructions — standing guidance you write once that steers the AI tools connected to it. Instead of re-explaining your world’s rules and tone in every session, you set them here, and they shape how agents work in that world.

Pick a starting point for a new world

You don’t have to write your world instructions from scratch. When you create a world, Orbismo starts it with a starting point — a ready-made set of world instructions that a connected AI reads to understand what the world is for and how to help you build it.

A Finish setup prompt on the new world’s dashboard opens the Set up your world page, where you choose that starting point. You can either:

• Use the default starter instructions (recommended) — a general onboarding brief. The first time you connect an AI tool, it walks you through what the world is for — asking one question at a time rather than making you fill out a form — starts capturing the people, places, and things you mention as you talk, and then offers to rewrite the instructions to fit how you’ll actually use the world.
• Or choose one of the ready-made templates — each geared to a particular kind of world, some seeding a little starter content to build on.

Either choice just fills in the world’s instructions (and any starter entities); nothing is locked in — you can edit or replace all of it later under Settings → World instructions.

More starting points live in the Orbismo Playbook — world templates that pair instructions with starter entities, and mini apps you can add to an existing world. No import tool needed: share a template’s files with a connected AI and ask it to set your world up.

When an AI offers to tailor the instructions, it only rewrites them after you’ve seen the draft and said yes (using the update_world_instructions tool). After that, the starter text is gone and your world’s own working instructions are in its place — which you can keep refining any time, by hand or by asking the AI.

What to put in it

Anything you’d otherwise have to repeat to an AI each time, for example:

• The world’s premise and tone (“a personal journal — write about people warmly and factually”).
• Naming and style conventions (“record people as ‘First Last’; write dates as YYYY-MM-DD”).
• Canon rules and boundaries (“never create an event without a confirmed date”).
• What to prioritize or avoid when creating or editing content.

Keep it focused — a tight set of guidelines works better than a sprawling one. For ready-to-paste starting points, see World instructions examples — or a complete example in Example: a journaling world.

Keep it short with Rules

For detailed or situational instructions, don’t inflate the world instructions — write them as Rule entities and leave only a trigger line here:

When the user mentions working out, look up the Workout Tracker rule and follow it.

The agent pulls the full instructions from the rule when the topic comes up. The world instructions stay a tight index of cues, and each behavior can be revised, paused, or retired by editing its rule — without touching the instructions.

Rules only work if the world instructions (or the user) point the agent at them — agents don’t scan for rules on their own. Either list a trigger line per rule, or add one standing line telling the agent to look up active rule entities at the start of each session. See Rules.

Setting it

Open the world’s Settings → World instructions. The editor shows how much of the size limit you’ve used (you can write up to about 32 KB — far more than you’ll usually need).

Or have your AI write them

You don’t have to write the instructions by hand. Connected agents get an update_world_instructions tool, so you can ask the AI to draft or revise them for you — “write world instructions for a journaling world,” or “add a line to my world instructions that dates are always ISO 8601” — and review the result under Settings → World instructions. The tool follows the same permissions as the editor: the world owner can always update, and editors can when the toggle below is on.

Who can edit it

• The owner can always edit the world instructions.
• The owner can let editors edit it too, via a toggle on the same screen. When that’s off, editors can read the instructions but not change them — in the portal and over MCP alike.
• Viewers can read it but not edit.

See Collaborators and permissions for how roles work.

How AI tools use it

Once you’ve connected a tool (via the connector or an API key), agents can load your world’s instructions so their work stays consistent with your canon — without you pasting the same guidance into every conversation.