Rool SDK

v0.3.12

The TypeScript SDK for Rool, a persistent and collaborative environment for organizing objects.

Building a new Rool extension? Start with @rool-dev/extension — it handles hosting, dev server, and gives you a reactive Svelte channel out of the box. This SDK is for advanced use cases: integrating Rool into an existing application, building Node.js scripts, or working outside the extension sandbox.

The SDK manages authentication, real-time synchronization, and media storage. Core primitives:

Spaces — Containers for objects, schema, metadata, and channels
Channels — Named contexts within a space. All object and AI operations go through a channel.
Conversations — Independent interaction histories within a channel.
Objects — Key-value records with any fields you define. References between objects are data fields whose values are object IDs.
AI operations — Create, update, or query objects using natural language and {{placeholders}}

Installation

npm install @rool-dev/sdk

Quick Start

import { RoolClient } from '@rool-dev/sdk';

const client = new RoolClient();
const authenticated = await client.initialize();

if (!authenticated) {
  client.login('My App');  // Redirects to auth page, shows "Sign in to My App"
}

// Create a new space, then open a channel on it
const space = await client.createSpace('Solar System');
const channel = await space.openChannel('main');

// Define the schema — what types of objects exist and their fields
await channel.createCollection('body', [
  { name: 'name', type: { kind: 'string' } },
  { name: 'mass', type: { kind: 'string' } },
  { name: 'radius', type: { kind: 'string' } },
  { name: 'orbits', type: { kind: 'maybe', inner: { kind: 'ref' } } },
]);

// Create objects with AI-generated content using {{placeholders}}
const { object: sun } = await channel.createObject({
  data: {
    name: 'Sun',
    mass: '{{mass in solar masses}}',
    radius: '{{radius in km}}'
  }
});

const { object: earth } = await channel.createObject({
  data: {
    name: 'Earth',
    mass: '{{mass in Earth masses}}',
    radius: '{{radius in km}}',
    orbits: sun.id  // Reference to the sun object
  }
});

// Use the AI agent to work with your data
const { message, objects } = await channel.prompt(
  'Add the other planets in our solar system, each referencing the Sun'
);
console.log(message);  // AI explains what it did
console.log(`Created ${objects.length} objects`);

// Query with natural language
const { objects: innerPlanets } = await channel.findObjects({
  prompt: 'planets closer to the sun than Earth'
});

// Clean up
channel.close();

Core Concepts

Spaces and Channels

A space is a container that holds objects, schema, metadata, and channels. A channel is a named context within a space — it’s the handle you use for all object and AI operations. Each channel contains one or more conversations, each with independent interaction history.

There are two main handles:

RoolSpace — Lightweight admin handle for user management, link access, channel management, and export. No real-time subscription.
RoolChannel — Full real-time handle for objects, AI prompts, media, schema, and undo/redo.

// Open a space for admin operations
const space = await client.openSpace('space-id');
await space.addUser(userId, 'editor');
await space.setLinkAccess('viewer');

// Open a channel for object and AI operations
const channel = await client.openChannel('space-id', 'my-channel');
await channel.prompt('Create some planets');

// Or open a channel via the space handle
const channel2 = await space.openChannel('research');
await channel2.prompt('Analyze the data');  // Independent channel

The channelId is fixed when you open a channel and cannot be changed. To use a different channel, open a new one. Channels to the same space share the same objects and schema.

Channel ID constraints:

1–32 characters
Only alphanumeric characters, hyphens (-), and underscores (_)

Conversations

A conversation is a named interaction history within a channel. By default, all operations use the 'default' conversation — most apps never need to think about conversations at all.

For apps that need multiple independent interaction threads (e.g., a chat sidebar with multiple threads), use channel.conversation() to get a handle scoped to a specific conversation:

// Default conversation — most apps use this
const channel = await client.openChannel('space-id', 'main');
await channel.prompt('Hello');  // Uses 'default' conversation

// Conversation handle — for multi-thread UIs
const thread = channel.conversation('thread-42');
await thread.prompt('Hello');  // Uses 'thread-42' conversation
thread.getInteractions();      // Interactions for thread-42 only

Each conversation has its own interaction history and optional system instruction. Conversations are auto-created on first interaction — no explicit create step needed. The 200-interaction cap applies per conversation. All conversations share one SSE connection per channel.

// System instructions are per-conversation
const thread = channel.conversation('research');
await thread.setSystemInstruction('Respond in haiku');

// List all conversations in this channel
const conversations = channel.getConversations();

// Delete a conversation (cannot delete 'default')
await channel.deleteConversation('old-thread');

// Rename a conversation
await thread.rename('Research Thread');

Branching Conversations

The conversation history is a tree, not a flat list. Each interaction has a parentId pointing to the interaction it continues from. When you call prompt(), the SDK automatically continues from the current active leaf. To branch (edit/reroll), pass a different parentInteractionId:

const thread = channel.conversation('chat');

// Normal conversation — each prompt auto-continues from the last
await thread.prompt('My favorite color is blue. Say OK.');
await thread.prompt('What is my favorite color?');  // Sees "blue"

// Branch: go back to the first message and say something different
const firstLeaf = thread.activeLeafId;  // ID of the "blue" interaction
const tree = thread.getTree();
const firstInteractionId = tree[firstLeaf!].parentId!;  // The root

await thread.prompt('My favorite color is red. Say OK.', {
  parentInteractionId: firstInteractionId,  // Sibling of "blue"
});
await thread.prompt('What is my favorite color?');  // Sees "red", not "blue"

// Switch back to the blue branch
thread.setActiveLeaf(firstLeaf!);
thread.getInteractions();  // Returns the blue branch (root → leaf)

Key concepts:

getInteractions() returns the active branch as a flat Interaction[] (root → leaf)
getTree() returns the full Record<string, Interaction> for branch navigation UI
activeLeafId is the tip of the branch the user is currently viewing
setActiveLeaf(id) switches branches (emits conversationUpdated so reactive UIs refresh)
prompt() with no parentInteractionId auto-continues from activeLeafId
prompt() with parentInteractionId: null starts a new root-level branch

Objects & References

Objects are plain key-value records. The id field is reserved; everything else is application-defined.

{ id: 'abc123', title: 'Hello World', status: 'draft' }

References between objects are data fields whose values are object IDs. The system detects these statistically — any string field whose value matches an existing object ID is recognized as a reference.

// A planet references a star via the 'orbits' field
{ id: 'earth', name: 'Earth', orbits: 'sun-01' }

// An array of references
{ id: 'team-a', name: 'Alpha', members: ['user-1', 'user-2', 'user-3'] }

References are just data — no special API is needed to create or remove them. Set a field to an object ID to create a reference; clear it to remove it.

AI Placeholder Pattern

Use {{description}} in field values to have AI generate content:

// Create with AI-generated content
await channel.createObject({
  data: {
    headline: '{{catchy headline about coffee}}',
    body: '{{informative paragraph}}'
  }
});

// Update existing content with AI
await channel.updateObject('abc123', {
  prompt: 'Make the body shorter and more casual'
});

// Add new AI-generated field to existing object
await channel.updateObject('abc123', {
  data: { summary: '{{one-sentence summary}}' }
});

When resolving placeholders, the agent has access to the full object data and the surrounding space context (except for _-prefixed fields). Placeholders are instructions, not templates, and do not need to repeat information already present in other fields.

Placeholders are resolved by the AI during the mutation and replaced with concrete values. The {{...}} syntax is never stored — it only guides the agent while creating or updating the object.

Checkpoints & Undo/Redo

Undo/redo works on checkpoints, not individual operations. Call checkpoint() before making changes to create a restore point.

// Create a checkpoint before user action
await channel.checkpoint('Delete object');
await channel.deleteObjects([objectId]);

// User can now undo back to the checkpoint
if (await channel.canUndo()) {
  await channel.undo(); // Restores the deleted object
}

// Redo reapplies the undone action
if (await channel.canRedo()) {
  await channel.redo(); // Deletes the object again
}

Without a checkpoint, undo() has nothing to restore to. Undo always restores the space to the last checkpoint, regardless of how many changes were made since.

In collaborative scenarios, conflicting changes (modified by others since your checkpoint) are silently skipped.

Hidden Fields

Fields starting with _ (e.g., _ui, _cache) are hidden from AI and ignored by the schema — you can add them to any object regardless of its collection definition. Otherwise they behave like normal fields: they sync in real-time, persist to the server, support undo/redo, and are visible to all users of the Space. Use them for UI state, positions, or other data the AI shouldn’t see or modify:

await channel.createObject({
  data: {
    title: 'My Article',
    author: "John Doe",
    _ui: { x: 100, y: 200, collapsed: false }
  }
});

Real-time Sync

Events fire for both local and remote changes. The source field indicates origin:

local_user — This client made the change
remote_user — Another user/client made the change
remote_agent — AI agent made the change
system — Resync after error

// All UI updates happen in one place, regardless of change source
channel.on('objectUpdated', ({ objectId, object, source }) => {
  renderObject(objectId, object);
  if (source === 'remote_agent') {
    doLayout(); // AI might have added content
  }
});

// Caller just makes the change - event handler does the UI work
channel.updateObject(objectId, { prompt: 'expand this' });

Custom Object IDs

By default, createObject generates a 6-character alphanumeric ID. Provide your own via data.id:

await channel.createObject({ data: { id: 'article-42', title: 'The Meaning of Life' } });

Why use custom IDs?

Fire-and-forget creation — Know the ID immediately without awaiting the response.
Meaningful IDs — Use domain-specific IDs like user-123 or doc-abc for easier debugging and external references.

// Fire-and-forget: create and reference without waiting
const id = RoolClient.generateId();
channel.createObject({ data: { id, text: '{{expand this idea}}' } });
channel.updateObject(parentId, { data: { notes: [...existingNotes, id] } }); // Add reference immediately

Constraints:

Must contain only alphanumeric characters, hyphens (-), and underscores (_)
Must be unique within the space (throws if ID exists)
Cannot be changed after creation (immutable)

Use RoolClient.generateId() when you need an ID before calling createObject but don’t need it to be meaningful — it gives you a valid random ID without writing your own generator.

Authentication

Browser (Default)

No configuration needed. Uses localStorage for tokens, redirects to login page.

const client = new RoolClient();
const authenticated = await client.initialize();

if (!authenticated) {
  client.login('My App'); // Redirect to the auth page
}

Node.js

For CLI tools and scripts. Stores credentials in ~/.config/rool/, opens browser for login.

import { NodeAuthProvider } from '@rool-dev/sdk/node';

const client = new RoolClient({ authProvider: new NodeAuthProvider() });
const authenticated = await client.initialize();

if (!authenticated) {
  await client.login('My CLI Tool'); // Opens browser, waits for callback
}

Auth Methods

Method	Description
`initialize(): Promise<boolean>`	Call on app startup. Processes auth callback from URL, sets up token refresh, returns auth state. Returns `false` if not authenticated. Throws if authenticated but account fetch fails (e.g. network error or invalid token).
`login(appName): void`	Redirect to login page. The app name is displayed on the auth page (“Sign in to {appName}”).
`logout(): void`	Clear tokens and state
`isAuthenticated(): Promise<boolean>`	Check auth status (validates token)
`getAuthUser(): AuthUser`	Get auth identity from JWT (`{ email, name }`)

AI Agent

The prompt() method is the primary way to invoke the AI agent. The agent has editor-level capabilities — it can create, modify, and delete objects — but cannot see or modify _-prefixed fields.

const { message, objects } = await channel.prompt(
  "Create a topic node for the solar system, then child nodes for each planet."
);
console.log(`AI: ${message}`);
console.log(`Modified ${objects.length} objects:`, objects);

Use checkpoint() before prompting to make operations undoable.

Method Signature

prompt(text: string, options?: PromptOptions): Promise<{ message: string; objects: RoolObject[] }>

Returns a message (the AI’s response) and the list of objects that were created or modified.

Options

Option	Description
`objectIds`	Focus the AI on specific objects (given primary attention in context)
`responseSchema`	Request structured JSON instead of text summary
`effort`	Effort level: `'QUICK'`, `'STANDARD'` (default), `'REASONING'`, or `'RESEARCH'`
`ephemeral`	If true, don’t record in interaction history (useful for tab completion)
`readOnly`	If true, disable mutation tools (create, update, delete). Use for questions.
`parentInteractionId`	Parent interaction in the conversation tree. Omit to auto-continue from the active leaf. Pass `null` to start a new root-level branch. Pass a specific ID to branch from that point (edit/reroll).
`attachments`	Files to attach (`File`, `Blob`, or `{ data, contentType }`). Uploaded to the media store via `uploadMedia()`. Resulting URLs are stored on the interaction’s `attachments` field for UI rendering. The AI can interpret images (JPEG, PNG, GIF, WebP, SVG), PDFs, text-based files (plain text, Markdown, CSV, HTML, XML, JSON), and DOCX documents. Other file types are uploaded and stored but the AI cannot read their contents.

Effort Levels

Level	Description
`QUICK`	Fast, lightweight model. Best for simple questions.
`STANDARD`	Default behavior with balanced capabilities.
`REASONING`	Extended reasoning for complex tasks.
`RESEARCH`	Most thorough mode with deep analysis. Slowest and most credit-intensive.

Examples

// Reorganize existing objects
const { objects } = await channel.prompt(
  "Group these notes by topic and create a parent node for each group."
);

// Work with specific objects
const result = await channel.prompt(
  "Summarize these articles",
  { objectIds: ['article-1', 'article-2'] }
);

// Quick question without mutations (fast model + read-only)
const { message } = await channel.prompt(
  "What topics are covered?",
  { effort: 'QUICK', readOnly: true }
);

// Complex analysis with extended reasoning
await channel.prompt(
  "Analyze relationships and reorganize",
  { effort: 'REASONING' }
);

// Attach files for the AI to see (File from <input>, Blob, or base64)
const file = fileInput.files[0]; // from <input type="file">
await channel.prompt(
  "Describe what's in this photo and create an object for it",
  { attachments: [file] }
);

Structured Responses

Use responseSchema to get structured JSON instead of a text message:

const { message } = await channel.prompt("Categorize these items", {
  objectIds: ['item-1', 'item-2', 'item-3'],
  responseSchema: {
    type: 'object',
    properties: {
      categories: {
        type: 'array',
        items: { type: 'string' }
      },
      summary: { type: 'string' }
    }
  }
});

const result = JSON.parse(message);
console.log(result.categories, result.summary);

Context Flow

AI operations automatically receive context:

Interaction history — Previous interactions and their results from this channel
Recently modified objects — Objects created or changed recently
Selected objects — Objects passed via objectIds are given primary focus

This context flows automatically — no configuration needed. The AI sees enough history to maintain coherent interactions while respecting the _-prefixed field hiding rules.

Collaboration

Adding Users to a Space

To add a user to a space, you need their user ID. Use searchUser() to find them by email:

// Find the user by email
const user = await client.searchUser('colleague@example.com');
if (!user) {
  throw new Error('User not found');
}

// Add them to the space
const space = await client.openSpace('space-id');
await space.addUser(user.id, 'editor');

Roles

Role	Capabilities
`owner`	Full control, can delete space and manage all users
`admin`	All editor capabilities, plus can manage users (except other admins/owners)
`editor`	Can create, modify, and delete objects
`viewer`	Read-only access (can query with `prompt` and `findObjects`)

Space Collaboration Methods

These methods are available on RoolSpace:

Method	Description
`listUsers(): Promise<SpaceMember[]>`	List users with access
`addUser(userId, role): Promise<void>`	Add user to space (requires owner or admin role)
`removeUser(userId): Promise<void>`	Remove user from space (requires owner or admin role)
`setLinkAccess(linkAccess): Promise<void>`	Set URL sharing level (requires owner or admin role)

Enable public URL access to allow anyone with the space URL to access it:

const space = await client.openSpace('space-id');

// Allow anyone with the URL to view
await space.setLinkAccess('viewer');

// Allow anyone with the URL to edit
await space.setLinkAccess('editor');

// Disable URL access (default)
await space.setLinkAccess('none');

// Check current setting
console.log(space.linkAccess); // 'none' | 'viewer' | 'editor'

When a user accesses a space via URL, they’re granted the corresponding role (viewer or editor) based on the space’s linkAccess setting.

Client User Methods

Method	Description
`currentUser: CurrentUser \| null`	Cached user profile from `initialize()`. Use for sync access to user info (id, email, name, etc.). Returns `null` before `initialize()` is called.
`getCurrentUser(): Promise<CurrentUser>`	Fetch fresh user profile from server (id, email, name, slug, plan, creditsBalance, totalCreditsUsed, createdAt, lastActivity, processedAt, storage)
`updateCurrentUser(input): Promise<CurrentUser>`	Update the current user’s profile (`name`, `slug`). Returns the updated user. Slug must be 3–32 chars, start with a letter, and contain only lowercase alphanumeric characters, hyphens, and underscores.
`searchUser(email): Promise<UserResult \| null>`	Find user by exact email address (no partial matching)

Real-time Collaboration

When multiple users have a space open, changes sync in real-time. The source field in events tells you who made the change:

channel.on('objectUpdated', ({ objectId, object, source }) => {
  if (source === 'remote_user') {
    // Another user made this change
    showCollaboratorActivity(object);
  }
});

See Real-time Sync for more on event sources.

RoolClient API

Logging

By default the SDK logs errors to the console. Pass a logger to see more or customize output:

// Default — errors only
const client = new RoolClient();

// Log everything to console
const client = new RoolClient({ logger: console });

// Bring your own logger (pino, winston, etc.)
const client = new RoolClient({
  logger: myLogger // any object with { debug, info, warn, error }
});

Space & Channel Lifecycle

Method	Description
`listSpaces(): Promise<RoolSpaceInfo[]>`	List available spaces
`openSpace(spaceId): Promise<RoolSpace>`	Open a space for admin operations (no real-time subscription)
`openChannel(spaceId, channelId): Promise<RoolChannel>`	Open a channel on a space
`createSpace(name): Promise<RoolSpace>`	Create a new space, returns admin handle
`deleteSpace(id): Promise<void>`	Permanently delete a space (cannot be undone)
`importArchive(name, archive): Promise<RoolSpace>`	Import from a zip archive, creating a new space

Channel Management

Manage channels within a space. Available on both the client and space handles:

Method	Description
`client.renameChannel(spaceId, channelId, name): Promise<void>`	Rename a channel
`client.deleteChannel(spaceId, channelId): Promise<void>`	Delete a channel and its interaction history
`space.getChannels(): ChannelInfo[]`	List channels (from cached snapshot)
`space.deleteChannel(channelId): Promise<void>`	Delete a channel
`channel.rename(name): Promise<void>`	Rename the current channel

User Storage

Server-side key-value storage for user preferences, UI state, and other persistent data. Replaces browser localStorage with cross-device, server-synced storage.

Features:

Fresh data fetched from server on initialize() — cache is authoritative after init
Sync reads from local cache (fast, no network round-trip)
Automatic sync to server and across tabs/devices via SSE
userStorageChanged event fires on all changes (local or remote)
Total storage limited to 10MB per user

Method	Description
`getUserStorage<T>(key): T \| undefined`	Get a value (sync, from cache)
`setUserStorage(key, value): void`	Set a value (updates cache, syncs to server)
`getAllUserStorage(): Record<string, unknown>`	Get all stored data (sync, from cache)

// After initialize(), storage is fresh from server
const authenticated = await client.initialize();

// Sync reads are now trustworthy
const theme = client.getUserStorage<string>('theme');
applyTheme(theme ?? 'light');

// Write - updates immediately, syncs to server in background
client.setUserStorage('theme', 'dark');
client.setUserStorage('sidebar', { collapsed: true, width: 280 });

// Delete a key
client.setUserStorage('theme', null);

// Listen for changes from other tabs/devices
client.on('userStorageChanged', ({ key, value, source }) => {
  // source: 'local' (this client) or 'remote' (server/other client)
  if (key === 'theme') applyTheme(value as string);
});

Extensions

Publish, manage, and discover extensions. See @rool-dev/extension for building extensions.

Method	Description
`publishExtension(extensionId, options): Promise<PublishedExtensionInfo>`	Publish or update an extension (`options.bundle`: zip with `index.html` and `manifest.json`)
`unpublishExtension(extensionId): Promise<void>`	Unpublish an extension
`listExtensions(): Promise<PublishedExtensionInfo[]>`	List your own published extensions
`getExtensionInfo(extensionId): Promise<PublishedExtensionInfo \| null>`	Get info for a specific extension
`findExtensions(options?): Promise<PublishedExtensionInfo[]>`	Search public extensions. Options: `query` (semantic search string), `limit` (default 20, max 100). Omit `query` to browse all.
`installExtension(spaceId, extensionId, channelId?): Promise<string>`	Install an extension into a space. Creates/updates a channel with the extension’s manifest settings (name, system instruction, collections). Returns the channel ID. Defaults `channelId` to `extensionId`.

Utilities

Method	Description
`RoolClient.generateId(): string`	Generate 6-char alphanumeric ID (static)
`graphql<T>(query, variables?): Promise<T>`	Execute raw GraphQL
`fetch(path, init?): Promise<Response>`	Authenticated fetch to the Rool API (adds Bearer token)
`destroy(): void`	Clean up resources

Client Events

client.on('authStateChanged', (authenticated: boolean) => void)
client.on('spaceAdded', (space: RoolSpaceInfo) => void)      // Space created or access granted
client.on('spaceRemoved', (spaceId: string) => void)         // Space deleted or access revoked
client.on('spaceRenamed', (spaceId: string, newName: string) => void)
client.on('channelCreated', (spaceId: string, channel: ChannelInfo) => void)
client.on('channelRenamed', (spaceId: string, channelId: string, newName: string) => void)
client.on('channelDeleted', (spaceId: string, channelId: string) => void)
client.on('userStorageChanged', ({ key, value, source }: UserStorageChangedEvent) => void)
client.on('connectionStateChanged', (state: 'connected' | 'disconnected' | 'reconnecting') => void)
client.on('error', (error: Error, context?: string) => void)

Space list management pattern:

const spaces = new Map<string, RoolSpaceInfo>();

client.on('spaceAdded', (space) => spaces.set(space.id, space));
client.on('spaceRemoved', (id) => spaces.delete(id));
client.on('spaceRenamed', (id, name) => {
  const space = spaces.get(id);
  if (space) spaces.set(id, { ...space, name });
});

RoolSpace API

A space is a lightweight admin handle for space-level operations. It does not have a real-time subscription — use channels for live data and object operations.

Properties

Property	Description
`id: string`	Space ID
`name: string`	Space name
`role: RoolUserRole`	User’s role
`linkAccess: LinkAccess`	URL sharing level
`memberCount: number`	Number of users with access to the space

Methods

Method	Description
`openChannel(channelId): Promise<RoolChannel>`	Open a channel on this space
`rename(newName): Promise<void>`	Rename this space
`delete(): Promise<void>`	Permanently delete this space
`listUsers(): Promise<SpaceMember[]>`	List users with access
`addUser(userId, role): Promise<void>`	Add user to space
`removeUser(userId): Promise<void>`	Remove user from space
`setLinkAccess(linkAccess): Promise<void>`	Set URL sharing level
`getChannels(): ChannelInfo[]`	List channels (from cached snapshot)
`deleteChannel(channelId): Promise<void>`	Delete a channel
`exportArchive(): Promise<Blob>`	Export space as zip archive
`refresh(): Promise<void>`	Refresh space data from server

RoolChannel API

A channel is a named context within a space. All object operations, AI prompts, and real-time sync go through a channel. The channelId is fixed at open time — to use a different channel, open a new one.

Properties

Property	Description
`id: string`	Space ID
`name: string`	Space name
`role: RoolUserRole`	User’s role (`'owner' \| 'admin' \| 'editor' \| 'viewer'`)
`linkAccess: LinkAccess`	URL sharing level (`'none' \| 'viewer' \| 'editor'`)
`userId: string`	Current user’s ID
`channelId: string`	Channel ID (read-only, fixed at open time)
`isReadOnly: boolean`	True if viewer role
`extensionUrl: string \| null`	URL of the installed extension, or null if this is a plain channel

Lifecycle

Method	Description
`close(): void`	Clean up resources and stop receiving updates
`rename(name): Promise<void>`	Rename this channel
`conversation(conversationId): ConversationHandle`	Get a handle scoped to a specific conversation (see Conversations)

Object Operations

Objects are plain key/value records. id is the only reserved field; everything else is application-defined. References between objects are data fields whose values are object IDs. All objects must belong to a collection (see below in the schema section). Before adding a new type of object, update the schema in the space.

Method	Description
`getObject(objectId): Promise<RoolObject \| undefined>`	Get object data, or undefined if not found.
`stat(objectId): RoolObjectStat \| undefined`	Get object stat (audit info: modifiedAt, modifiedBy, modifiedByName), or undefined if not found. Sync read from local cache.
`findObjects(options): Promise<{ objects, message }>`	Find objects using structured filters and natural language. Results sorted by modifiedAt (desc by default).
`getObjectIds(options?): string[]`	Get all object IDs. Sorted by modifiedAt (desc by default). Options: `{ limit?, order? }`.
`createObject(options): Promise<{ object, message }>`	Create a new object. Returns the object (with AI-filled content) and message.
`updateObject(objectId, options): Promise<{ object, message }>`	Update an existing object. Returns the updated object and message.
`deleteObjects(objectIds): Promise<void>`	Delete objects. Other objects referencing deleted objects retain stale ref values.

createObject Options

Option	Description
`data`	Object data fields (required). Include `id` to use a custom ID. Use `{{placeholder}}` for AI-generated content. Fields prefixed with `_` are hidden from AI.
`ephemeral`	If true, the operation won’t be recorded in interaction history. Useful for transient operations.

updateObject Options

Option	Description
`data`	Fields to add or update. Pass `null`/`undefined` to delete a field. Use `{{placeholder}}` for AI-generated content. Fields prefixed with `_` are hidden from AI.
`prompt`	Natural language instruction for AI to modify content.
`ephemeral`	If true, the operation won’t be recorded in interaction history. Useful for transient operations.

findObjects Options

Find objects using structured filters and/or natural language.

where only — exact-match filtering, no AI, no credits.
collection only — filter by collection name (shape-based matching), no AI, no credits.
prompt only — AI-powered semantic query over all objects.
where + prompt — where (and objectIds) narrow the data set first, then the AI queries within the constrained set.

Option	Description
`where`	Exact-match field filter (e.g. `{ status: 'published' }`). Values must match literally — no operators or `{{placeholders}}`. When combined with `prompt`, constrains which objects the AI can see.
`collection`	Filter by collection name. Only returns objects whose shape matches the named collection.
`prompt`	Natural language query. Triggers AI evaluation (uses credits).
`limit`	Maximum number of results.
`objectIds`	Scope to specific object IDs. Constrains the candidate set in both structured and AI queries.
`order`	Sort order by modifiedAt: `'asc'` or `'desc'` (default: `'desc'`).
`ephemeral`	If true, the query won’t be recorded in interaction history. Useful for responsive search.

Examples:

// Filter by collection (no AI, no credits)
const { objects } = await channel.findObjects({
  collection: 'article'
});

// Exact field matching (no AI, no credits)
const { objects } = await channel.findObjects({
  where: { status: 'published' }
});

// Combine collection and field filters
const { objects } = await channel.findObjects({
  collection: 'article',
  where: { status: 'published' }
});

// Pure natural language query (AI interprets)
const { objects, message } = await channel.findObjects({
  prompt: 'articles about space exploration published this year'
});

// Combined: collection + where narrow the data, prompt queries within it
const { objects } = await channel.findObjects({
  collection: 'article',
  prompt: 'that discuss climate solutions positively',
  limit: 10
});

When where or objectIds are provided with a prompt, the AI only sees the filtered subset — not the full space. The returned message explains the query result.

Undo/Redo

Method	Description
`checkpoint(label?): Promise<string>`	Call before mutations. Saves current state for undo.
`canUndo(): Promise<boolean>`	Check if undo available
`canRedo(): Promise<boolean>`	Check if redo available
`undo(): Promise<boolean>`	Undo to previous checkpoint
`redo(): Promise<boolean>`	Redo undone action
`clearHistory(): Promise<void>`	Clear undo/redo stack

See Checkpoints & Undo/Redo for semantics.

Space Metadata

Store arbitrary data alongside the Space without it being part of the object data (e.g., viewport state, user preferences).

Method	Description
`setMetadata(key, value): void`	Set space-level metadata
`getMetadata(key): unknown`	Get metadata value, or undefined if key not set
`getAllMetadata(): Record<string, unknown>`	Get all metadata

Media

Media URLs in object fields are visible to AI. Both uploaded and AI-generated media work the same way — use fetchMedia to retrieve them for display.

Method	Description
`uploadMedia(file): Promise<string>`	Upload file, returns URL
`fetchMedia(url, options?): Promise<MediaResponse>`	Fetch any URL, returns headers and blob() method (adds auth for backend URLs, works for external URLs too). Pass `{ forceProxy: true }` to skip the direct fetch and route through the server proxy immediately.
`deleteMedia(url): Promise<void>`	Delete media file by URL
`listMedia(): Promise<MediaInfo[]>`	List all media with metadata

// Upload an image
const url = await channel.uploadMedia(file);
await channel.createObject({ data: { title: 'Photo', image: url } });

// Or let AI generate one using a placeholder
await channel.createObject({
  data: { title: 'Mascot', image: '{{generate an image of a flying tortoise}}' }
});

// Display media (handles auth automatically)
const response = await channel.fetchMedia(object.image);
if (response.contentType.startsWith('image/')) {
  const blob = await response.blob();
  img.src = URL.createObjectURL(blob);
}

Collection Schema

Collections are types you can use to group objects in a space. Every object must belong to a collection. Collections make up the schema and are stored in the space data, syncing in real time together with the rest of the space. The schema is also visible to the AI agent, which it can use to understand what collections exist and what fields they contain, producing more consistent objects.

// Define a collection with typed fields
await channel.createCollection('article', [
  { name: 'title', type: { kind: 'string' } },
  { name: 'status', type: { kind: 'enum', values: ['draft', 'published', 'archived'] } },
  { name: 'tags', type: { kind: 'array', inner: { kind: 'string' } } },
  { name: 'author', type: { kind: 'ref' } },
]);

// Read the current schema
const schema = channel.getSchema();
console.log(schema.article.fields); // FieldDef[]

// Modify an existing collection's fields
await channel.alterCollection('article', [
  { name: 'title', type: { kind: 'string' } },
  { name: 'status', type: { kind: 'enum', values: ['draft', 'review', 'published', 'archived'] } },
  { name: 'tags', type: { kind: 'array', inner: { kind: 'string' } } },
  { name: 'author', type: { kind: 'ref' } },
  { name: 'wordCount', type: { kind: 'number' } },
]);

// Remove a collection
await channel.dropCollection('article');

Method	Description
`getSchema(): SpaceSchema`	Get all collection definitions
`createCollection(name, fields): Promise<CollectionDef>`	Add a new collection to the schema
`alterCollection(name, fields): Promise<CollectionDef>`	Replace a collection’s field definitions
`dropCollection(name): Promise<void>`	Remove a collection from the schema

Field Types

Kind	Description	Example
`string`	Text value	`{ kind: 'string' }`
`number`	Numeric value	`{ kind: 'number' }`
`boolean`	True/false	`{ kind: 'boolean' }`
`ref`	Reference to another object	`{ kind: 'ref' }`
`enum`	One of a set of values	`{ kind: 'enum', values: ['a', 'b'] }`
`literal`	Exact value	`{ kind: 'literal', value: 'fixed' }`
`array`	List of values	`{ kind: 'array', inner: { kind: 'string' } }`
`maybe`	Optional (nullable)	`{ kind: 'maybe', inner: { kind: 'number' } }`

Import/Export

Export and import space data as zip archives for backup, portability, or migration:

Method	Description
`space.exportArchive(): Promise<Blob>`	Export objects, metadata, channels, and media as a zip archive
`client.importArchive(name, archive): Promise<RoolSpace>`	Import from a zip archive, creating a new space

Export:

const space = await client.openSpace('space-id');
const archive = await space.exportArchive();
// Save as .zip file
const url = URL.createObjectURL(archive);

Import:

const space = await client.importArchive('Imported Data', archiveBlob);
const channel = await space.openChannel('main');

The archive format bundles data.json (with objects, metadata, and channels) and a media/ folder containing all media files. Media URLs are rewritten to relative paths within the archive and restored on import.

Channel Events

Semantic events describe what changed. Events fire for both local changes and remote changes.

// source indicates origin:
// - 'local_user': This client made the change
// - 'remote_user': Another user/client made the change
// - 'remote_agent': AI agent made the change
// - 'system': Resync after error

// Object events
channel.on('objectCreated', ({ objectId, object, source }) => void)
channel.on('objectUpdated', ({ objectId, object, source }) => void)
channel.on('objectDeleted', ({ objectId, source }) => void)

// Space metadata
channel.on('metadataUpdated', ({ metadata, source }) => void)

// Collection schema changed
channel.on('schemaUpdated', ({ schema, source }) => void)

// Channel metadata updated (name, extensionUrl)
channel.on('channelUpdated', ({ channelId, source }) => void)

// Conversation interaction history updated
channel.on('conversationUpdated', ({ conversationId, channelId, source }) => void)

// Full state replacement (undo/redo, resync after error)
channel.on('reset', ({ source }) => void)

// Sync error occurred, channel resynced from server
channel.on('syncError', (error: Error) => void)

Error Handling

AI operations may fail due to rate limiting or other transient errors. Check error.message for user-friendly error text:

try {
  await channel.updateObject(objectId, { prompt: 'expand this' });
} catch (error) {
  if (error.message.includes('temporarily unavailable')) {
    showToast('Service busy, please try again in a moment');
  } else {
    showToast(error.message);
  }
}

Interaction History

Each channel contains one or more conversations, each with its own interaction history. History is stored as a tree (interactions linked by parentId) in the space data and syncs in real-time. Capped at 200 interactions per conversation.

Conversation History Methods

Method	Description
`getInteractions(): Interaction[]`	Get the active branch as a flat array (root → leaf)
`getTree(): Record<string, Interaction>`	Get the full interaction tree for branch navigation
`activeLeafId: string \| undefined`	The tip of the currently active branch
`setActiveLeaf(id: string): void`	Switch to a different branch (emits `conversationUpdated`)
`getSystemInstruction(): string \| undefined`	Get system instruction for the default conversation
`setSystemInstruction(instruction): Promise<void>`	Set system instruction for the default conversation. Pass `null` to clear.
`getConversations(): ConversationInfo[]`	List all conversations in this channel
`deleteConversation(conversationId): Promise<void>`	Delete a conversation (cannot delete `'default'`)
`renameConversation(name): Promise<void>`	Rename the default conversation

Channel management (listing, renaming, deleting channels) is done via the client — see Channel Management.

The ai Field

The ai field in interactions distinguishes AI-generated responses from synthetic confirmations:

ai: true — AI processed this operation (prompt, or createObject/updateObject with placeholders)
ai: false — System confirmation only (e.g., “Created object abc123”)

Tool Calls

The toolCalls array captures what the AI agent did during execution. The conversationUpdated event fires when each tool starts and completes. A tool call without a result is still running; once result is present, the tool has finished.

Data Types

Schema Types

// Allowed field types
type FieldType =
  | { kind: 'string' }
  | { kind: 'number' }
  | { kind: 'boolean' }
  | { kind: 'array'; inner?: FieldType }
  | { kind: 'maybe'; inner: FieldType }
  | { kind: 'enum'; values: string[] }
  | { kind: 'literal'; value: string | number | boolean }
  | { kind: 'ref' };

interface FieldDef {
  name: string;
  type: FieldType;
}

interface CollectionDef {
  fields: FieldDef[];
}

// Full schema — collection names to definitions
type SpaceSchema = Record<string, CollectionDef>;

Object Data

// RoolObject represents the object data you work with
// Always contains `id`, plus any additional fields
// Fields prefixed with _ are hidden from AI
// References between objects are fields whose values are object IDs
interface RoolObject {
  id: string;
  [key: string]: unknown;
}

// Object stat - audit information returned by channel.stat()
interface RoolObjectStat {
  modifiedAt: number;
  modifiedBy: string;
  modifiedByName: string | null;
}

Channels and Conversations

// Conversation — holds interaction tree and optional system instruction
interface Conversation {
  name?: string;                  // Conversation name (optional)
  systemInstruction?: string;     // Custom system instruction for AI
  createdAt: number;              // Timestamp when conversation was created
  createdBy: string;              // User ID who created the conversation
  interactions: Record<string, Interaction>;  // Interaction tree (keyed by ID, linked by parentId)
}

// Conversation summary info (returned by channel.getConversations())
interface ConversationInfo {
  id: string;
  name: string | null;
  systemInstruction: string | null;
  createdAt: number;
  createdBy: string;
  interactionCount: number;
}

// Channel container with metadata and conversations
interface Channel {
  name?: string;                // Channel name (optional)
  createdAt: number;            // Timestamp when channel was created
  createdBy: string;            // User ID who created the channel
  createdByName?: string;       // Display name at time of creation
  extensionUrl?: string;        // URL of installed extension (set by installExtension)
  conversations: Record<string, Conversation>;  // Keyed by conversation ID
}

// Channel summary info (returned by client.getChannels)
interface ChannelInfo {
  id: string;
  name: string | null;
  createdAt: number;
  createdBy: string;
  createdByName: string | null;
  interactionCount: number;
  extensionUrl: string | null;  // URL of installed extension, or null
}

Note: Channel and ChannelInfo are data types describing the stored channel metadata. The Channel interface is the wire format; RoolChannel is the live SDK class you interact with.

Interaction Types

interface ToolCall {
  name: string;      // Tool name (e.g., "create_object", "update_object", "search_web")
  input: unknown;    // Arguments passed to the tool
  result?: string;   // Truncated result (absent while tool is running)
}

type InteractionStatus = 'pending' | 'streaming' | 'done' | 'error';

interface Interaction {
  id: string;                    // Unique ID for this interaction
  parentId: string | null;       // Parent in conversation tree (null = root)
  timestamp: number;
  userId: string;                // Who performed this interaction
  userName?: string | null;      // Display name at time of interaction
  operation: 'prompt' | 'createObject' | 'updateObject' | 'deleteObjects';
  input: string;                 // What the user did: prompt text or action description
  output: string | null;         // AI response or confirmation message (may be partial when streaming)
  status: InteractionStatus;     // Lifecycle status (pending → streaming → done/error)
  ai: boolean;                   // Whether AI was invoked (vs synthetic confirmation)
  modifiedObjectIds: string[];   // Objects affected by this interaction
  toolCalls: ToolCall[];         // Tools called during this interaction (for AI prompts)
  attachments?: string[];        // Media URLs attached by the user (images, documents, etc.)
}

Info Types

type RoolUserRole = 'owner' | 'admin' | 'editor' | 'viewer';
type LinkAccess = 'none' | 'viewer' | 'editor';

interface RoolSpaceInfo { id: string; name: string; role: RoolUserRole; ownerId: string; size: number; createdAt: string; updatedAt: string; linkAccess: LinkAccess; memberCount: number; }
interface SpaceMember { id: string; email: string; role: RoolUserRole; }
interface UserResult { id: string; email: string; name: string | null; }
interface CurrentUser { id: string; email: string; name: string | null; slug: string; plan: string; creditsBalance: number; totalCreditsUsed: number; createdAt: string; lastActivity: string; processedAt: string; storage: Record<string, unknown>; }
interface MediaInfo { url: string; contentType: string; size: number; createdAt: string; }
interface MediaResponse { contentType: string; size: number | null; blob(): Promise<Blob>; }
type ChangeSource = 'local_user' | 'remote_user' | 'remote_agent' | 'system';

Prompt Options

type PromptEffort = 'QUICK' | 'STANDARD' | 'REASONING' | 'RESEARCH';

interface PromptOptions {
  objectIds?: string[];      // Scope to specific objects
  responseSchema?: Record<string, unknown>;
  effort?: PromptEffort;     // Effort level (default: 'STANDARD')
  ephemeral?: boolean;       // Don't record in interaction history
  readOnly?: boolean;        // Disable mutation tools (default: false)
  parentInteractionId?: string | null;  // Branch from a specific interaction (omit to auto-continue)
  attachments?: Array<File | Blob | { data: string; contentType: string }>;  // Files to attach (uploaded to media store)
}

License

MIT - see LICENSE for details.

Rool SDK

Installation

Quick Start

Core Concepts

Spaces and Channels

Conversations

Branching Conversations

Objects & References

AI Placeholder Pattern

Checkpoints & Undo/Redo

Hidden Fields

Real-time Sync

Custom Object IDs

Authentication

Browser (Default)

Node.js

Auth Methods

AI Agent

Method Signature

Options

Effort Levels

Examples

Structured Responses

Context Flow

Collaboration

Adding Users to a Space

Roles

Space Collaboration Methods

URL Sharing

Client User Methods

Real-time Collaboration

RoolClient API

Logging

Space & Channel Lifecycle

Channel Management

User Storage

Extensions

Utilities

Client Events

RoolSpace API

Properties

Methods

RoolChannel API

Properties

Lifecycle

Object Operations

createObject Options

updateObject Options

findObjects Options

Undo/Redo

Space Metadata

Media

Collection Schema

Field Types

Import/Export

Channel Events

Error Handling

Interaction History

Conversation History Methods

The ai Field

Tool Calls

Data Types

Schema Types

Object Data

Channels and Conversations

Interaction Types

Info Types

Prompt Options

License