@rool-dev/sdk 0.8.2-dev.02b70e5 → 0.8.2-dev.d82ea25

package/README.md

@@ -1,16 +1,14 @@
 # Rool SDK
 
-The TypeScript SDK for Rool, a persistent and collaborative environment for organizing objects.
+TypeScript SDK for Rool, a persistent collaborative workspace for objects, AI-assisted editing, and per-space files.
 
-> **Building a new Rool extension?** Start with [`@rool-dev/extension`](/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.
+Core primitives:
 
-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}}`
+- **Spaces** — containers for objects, schema, metadata, channels, collaborators, and files.
+- **Channels** — named contexts within a space for object operations and AI conversations.
+- **Conversations** — independent interaction histories within a channel.
+- **Objects** — JSON records addressed by object paths such as `/space/article/welcome.json`.
+- **Files** — user-visible files stored under `/rool-drive/...` through WebDAV.
 
 ## Installation
 
@@ -23,1074 +21,611 @@ npm install @rool-dev/sdk
 ```typescript
 import { RoolClient } from '@rool-dev/sdk';
 
-const client = new RoolClient();
-const authenticated = await client.initialize();
+async function main() {
+  const client = new RoolClient();
 
-if (!authenticated) {
-  client.login('My App');  // Redirects to auth page, shows "Sign in to My App"
-}
+  if (!(await client.initialize())) {
+    await client.login('My App');
+    // Browser auth redirects away. Run startup again after the auth callback.
+    return;
+  }
 
-// Create a new space, then open a channel on it
-const space = await client.createSpace('Solar System');
-const channel = await space.openChannel('main');
+  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' } } },
-]);
+  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: {
-    type: 'body',  // Must match a collection name
+  const { object: sun } = await channel.putObject('/space/body/sun.json', {
     name: 'Sun',
-    mass: '{{mass in solar masses}}',
-    radius: '{{radius in km}}'
-  }
-});
+    mass: '1 solar mass',
+    radius: '696,340 km',
+  });
 
-const { object: earth } = await channel.createObject({
-  data: {
-    type: 'body',
+  const { object: earth } = await channel.putObject('/space/body/earth.json', {
     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();
-```
+    mass: '1 Earth mass',
+    radius: '6,371 km',
+    orbits: sun.path,
+  });
 
-## Core Concepts
+  const { message, objects } = await channel.prompt(
+    'Add the other planets in our solar system, each referencing the Sun.'
+  );
 
-### 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`** — Live handle with SSE subscription for user management, link access, channel management, export, and channel lifecycle events. Extends `EventEmitter`.
-- **`RoolChannel`** — Full real-time handle for objects, AI prompts, media, schema, and undo/redo.
-
-```typescript
-// Open a space — live handle with SSE subscription
-const space = await client.openSpace('space-id');
-await space.addUser(userId, 'editor');
-await space.setLinkAccess('viewer');
+  console.log(message);
+  console.log(`Modified ${objects.length} objects`);
 
-// React to channel changes in real-time
-space.on('channelCreated', (channel) => console.log('New channel:', channel.id));
-space.on('channelUpdated', (channel) => console.log('Updated:', channel.id));
-space.on('channelDeleted', (channelId) => console.log('Deleted:', channelId));
+  const loadedEarth = await channel.getObject(earth.path);
+  console.log(loadedEarth?.body.name);
 
-// Open a channel for object and AI operations
-const channel = await space.openChannel('my-channel');
-await channel.prompt('Create some planets');
-
-// Open another channel on the same space
-const channel2 = await space.openChannel('research');
-await channel2.prompt('Analyze the data');  // Independent channel
+  space.close();
+}
 
-// Clean up — stops subscription and closes all open channels
-space.close();
+void main();
 ```
 
-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 (`_`)
+## Paths and Resource URIs
 
-### Conversations
+Most SDK methods take plain path strings:
 
-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:
-
-```typescript
-// Default conversation — most apps use this
-const space = await client.openSpace('space-id');
-const channel = await space.openChannel('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
-```
+- Object paths: `/space/<collection>/<name>.json` (exactly three segments; no dotfile collection or object names)
+- File paths: `/rool-drive/<path/to/file>`
 
-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.
+`rool-machine:/...` URIs are the user-facing/canonical form for resource references in prompt attachments and interaction history. The exported helpers normalize between these forms when you need them.
 
 ```typescript
-// System instructions are per-conversation
-const thread = channel.conversation('research');
-await thread.setSystemInstruction('Respond in haiku');
+import { machinePath, machineUri, isObjectPath } from '@rool-dev/sdk';
 
-// List all conversations in this channel
-const conversations = channel.getConversations();
+machinePath('rool-machine:/rool-drive/docs/read%20me.md');
+// '/rool-drive/docs/read me.md'
 
-// Delete a conversation (cannot delete 'default')
-await channel.deleteConversation('old-thread');
+machineUri('/space/article/welcome.json');
+// 'rool-machine:/space/article/welcome.json'
 
-// Rename a conversation
-await thread.rename('Research Thread');
+isObjectPath('/space/article/welcome.json'); // true
 ```
 
-### 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`:
+Object APIs require full object paths. References between objects are ordinary body fields containing object paths:
 
 ```typescript
-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)
+{
+  path: '/space/body/earth.json',
+  body: { name: 'Earth', orbits: '/space/body/sun.json' },
+}
 ```
 
-**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
+## Authentication
 
-### Objects & References
+### Browser
 
-**Objects** are plain key-value records. `id` and `type` are reserved; everything else is application-defined. Every object must include a `type` field whose value names a collection in the schema — that binds the object to that collection and determines how it's validated. Create the collection first (see [Collection Schema](#collection-schema)).
+The default auth provider stores tokens in browser storage and redirects to the Rool auth page.
 
 ```typescript
-{ id: 'abc123', type: 'article', 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.
+async function start() {
+  const client = new RoolClient();
 
-```typescript
-// A planet references a star via the 'orbits' field
-{ id: 'earth', type: 'body', name: 'Earth', orbits: 'sun-01' }
-
-// An array of references
-{ id: 'team-a', type: 'team', 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:
-
-```typescript
-// Create with AI-generated content
-await channel.createObject({
-  data: {
-    type: 'article',
-    headline: '{{catchy headline about coffee}}',
-    body: '{{informative paragraph}}'
+  if (!(await client.initialize())) {
+    await client.login('My App');
+    // Browser auth redirects; stop startup until the callback reloads the app.
+    return;
   }
-});
 
-// Update existing content with AI
-await channel.updateObject('abc123', {
-  prompt: 'Make the body shorter and more casual'
-});
+  // Use the authenticated client here.
+}
 
-// Add new AI-generated field to existing object
-await channel.updateObject('abc123', {
-  data: { summary: '{{one-sentence summary}}' }
-});
+void start();
 ```
 
-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
+### Node.js
 
-Undo/redo works on **checkpoints**, not individual operations. Call `checkpoint()` before making changes to create a restore point. Each checkpoint stores a snapshot of the entire space.
+Use the Node auth provider for CLIs and scripts. It stores endpoint-scoped credentials under `~/.config/rool/` by default (for example, `credentials-<hash>.json`) and opens a browser for login.
 
 ```typescript
-// Create a checkpoint before user action
-await channel.checkpoint('Delete object');
-await channel.deleteObjects([objectId]);
+import { RoolClient } from '@rool-dev/sdk';
+import { NodeAuthProvider } from '@rool-dev/sdk/node';
 
-// User can now undo back to the checkpoint
-if (await channel.canUndo()) {
-  await channel.undo(); // Restores the deleted object
-}
+const client = new RoolClient({ authProvider: new NodeAuthProvider() });
+let authenticated = await client.initialize();
 
-// Redo reapplies the undone action
-if (await channel.canRedo()) {
-  await channel.redo(); // Deletes the object again
+if (!authenticated) {
+  await client.login('My CLI Tool');
+  // Re-run initialize after the non-redirect login to hydrate currentUser,
+  // user storage, and client-level event subscriptions.
+  authenticated = await client.initialize();
 }
-```
-
-Checkpoints are **space-wide**: one shared stack across all channels and users. `undo()` restores the entire space — including any work others did since the checkpoint. Stacks are capped at 25 entries; identical-content checkpoints are deduped; a new checkpoint clears the redo stack.
-
-### 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:
-
-```typescript
-await channel.createObject({
-  data: {
-    type: 'article',
-    title: 'My Article',
-    author: "John Doe",
-    _ui: { x: 100, y: 200, collapsed: false }
-  }
-});
+if (!authenticated) throw new Error('Login required');
 ```
 
-### 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
+### Auth API
 
-```typescript
-// 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' });
-```
+| Method | Description |
+| --- | --- |
+| `initialize(): Promise<boolean>` | Call on startup. Initializes auth, refreshes user/storage state, and starts client events when authenticated. |
+| `login(appName, params?): Promise<void>` | Start login flow. |
+| `signup(appName, params?): Promise<void>` | Start signup flow. |
+| `verify(token): Promise<boolean>` | Complete email verification token flow; returns `false` when the active auth provider does not implement verification. |
+| `logout(): void` | Clear auth state and close open spaces. |
+| `isAuthenticated(): Promise<boolean>` | Validate current auth state. |
+| `getAuthUser(): AuthUser` | Return auth identity decoded from the token. |
+| `setPassword(password): Promise<void>` | Set/change password for the current user. |
 
-### Custom Object IDs
+## Spaces and Channels
 
-By default, `createObject` generates a 6-character alphanumeric ID. Provide your own via `data.id`:
+Open a space to receive live events and manage collaborators, file storage, and channels. Open a channel to work with objects, schema, metadata, and AI.
 
 ```typescript
-await channel.createObject({ data: { id: 'article-42', type: 'article', title: 'The Meaning of Life' } });
-```
+const space = await client.openSpace('space-id');
 
-**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.
+space.on('channelCreated', (channel) => console.log(channel.id));
+space.on('filesChanged', () => console.log('files changed'));
 
-```typescript
-// Fire-and-forget: create and reference without waiting
-const id = RoolClient.generateId();
-channel.createObject({ data: { id, type: 'note', text: '{{expand this idea}}' } });
-channel.updateObject(parentId, { data: { notes: [...existingNotes, id] } }); // Add reference immediately
+const channel = await space.openChannel('main');
+await channel.prompt('Summarize this space');
 ```
 
-**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)
+Channel IDs must be 1–32 characters and contain only letters, numbers, `_`, and `-`.
 
-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
+## Object Operations
 
-### Browser (Default)
-
-No configuration needed. Uses localStorage for tokens, redirects to login page.
+Objects are JSON files under `/space`. Create the collection before writing objects in it.
 
 ```typescript
-const client = new RoolClient();
-const authenticated = await client.initialize();
-
-if (!authenticated) {
-  client.login('My App'); // Redirect to the auth page
-}
-```
+await channel.createCollection('article', [
+  { name: 'title', type: { kind: 'string' } },
+  { name: 'status', type: { kind: 'string' } },
+]);
 
-### Node.js
+// Create or replace an exact object path
+const { object } = await channel.putObject('/space/article/welcome.json', {
+  title: 'Welcome',
+  status: 'draft',
+});
 
-For CLI tools and scripts. Stores credentials in `~/.config/rool/`, opens browser for login.
+// Patch fields; null or undefined deletes a field
+await channel.patchObject(object.path, {
+  data: { status: 'published', obsoleteField: null },
+});
 
-```typescript
-import { NodeAuthProvider } from '@rool-dev/sdk/node';
+// Read one or many objects
+await channel.getObject('/space/article/welcome.json');
+await channel.getObjects([
+  '/space/article/welcome.json',
+  '/space/article/intro.json',
+]);
 
-const client = new RoolClient({ authProvider: new NodeAuthProvider() });
-const authenticated = await client.initialize();
+// Rename or move an object
+await channel.moveObject(
+  '/space/article/welcome.json',
+  '/space/article/hello-world.json'
+);
 
-if (!authenticated) {
-  await client.login('My CLI Tool'); // Opens browser, waits for callback
-}
+// Delete objects
+await channel.deleteObjects(['/space/article/hello-world.json']);
 ```
 
-### 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, params?): void` | Redirect to login page. The app name is displayed on the auth page ("Sign in to {appName}"). Optional `params` are added as query parameters to the auth URL. |
-| `signup(appName, params?): void` | Redirect to signup page. The app name is displayed on the auth page ("Sign up for {appName}"). Optional `params` are added as query parameters to the auth URL. |
-| `verify(token): Promise<boolean>` | Sign in using a verification token (from a `?verify=<token>` email link). Used by the official Rool app — most integrations won't need this. |
-| `logout(): void` | Clear tokens and state |
-| `isAuthenticated(): Promise<boolean>` | Check auth status (validates token) |
-| `getAuthUser(): AuthUser` | Get auth identity from JWT (`{ email, name }`) |
-| `setPassword(password): Promise<void>` | Set or change the current user's password. Requires an authenticated session. Password must be at least 8 characters and contain both letters and either digits or symbols. Throws with a human-readable message on validation or server failure. |
+| --- | --- |
+| `getObject(path): Promise<RoolObject | undefined>` | Fetch one object by object path. |
+| `getObjects(paths): Promise<GetObjectsResult>` | Fetch objects in bulk; returns `objects` and `missing`. |
+| `stat(path): RoolObjectStat | undefined` | Cached audit info for an object. |
+| `putObject(path, body): Promise<{ object, message }>` | Create or replace an object at an exact path. |
+| `patchObject(path, { data }): Promise<{ object, message }>` | Patch an object's body; `null`/`undefined` deletes fields. |
+| `moveObject(from, to, options?): Promise<{ object, message }>` | Rename or relocate an object; `options.body` can replace the body after moving. |
+| `deleteObjects(paths): Promise<void>` | Delete object files. |
 
 ## 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.
+`prompt()` invokes the AI agent. The agent can inspect space context and, unless `readOnly` or a read-only effort is used, create/modify/move/delete objects.
 
 ```typescript
 const { message, objects } = await channel.prompt(
-  "Create a topic node for the solar system, then child nodes for each planet."
+  '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
-
-```typescript
-prompt(text: string, options?: PromptOptions): Promise<{ message: string; objects: RoolObject[] }>
+console.log(message);
+console.log(objects.map((object) => object.path));
 ```
 
-Returns a message (the AI's response) and the list of objects that were created or modified.
-
-### Options
+### Prompt 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
-
-```typescript
-// 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] }
-);
-```
+| --- | --- |
+| `responseSchema` | Request structured JSON text matching a JSON-schema-like shape. |
+| `effort` | `'QUICK'` (fast/read-only), `'STANDARD'` (default), `'REASONING'`, or `'RESEARCH'`. |
+| `ephemeral` | Do not record the prompt in interaction history. |
+| `readOnly` | Disable mutation tools. |
+| `parentInteractionId` | Conversation-tree parent. Omit to continue from the active leaf; pass `null` for a new root branch. |
+| `attachments` | Existing object/file paths or `rool-machine:/...` URIs, plus local files (`File`, `Blob`, or `{ data, contentType, filename? }`). |
+| `signal` | AbortSignal used to request that the server stop an in-flight prompt. |
+| `eventName` | Optional telemetry event name. Defaults to `'prompt_user'`. |
+
+```typescript
+// Read-only quick question
+await channel.prompt('What topics are covered?', {
+  effort: 'QUICK', // fast/read-only
+});
 
-### Structured Responses
+// Focus on existing objects and files
+await channel.prompt('Compare these resources', {
+  attachments: [
+    '/space/article/intro.json',
+    'rool-machine:/rool-drive/docs/report.pdf',
+  ],
+});
 
-Use `responseSchema` to get structured JSON instead of a text message:
+// Upload a local file as an attachment
+await channel.prompt('Describe this image', {
+  attachments: [fileInput.files![0]],
+});
 
-```typescript
-const { message } = await channel.prompt("Categorize these items", {
-  objectIds: ['item-1', 'item-2', 'item-3'],
+// Structured response
+const { message } = await channel.prompt('Categorize these items', {
   responseSchema: {
     type: 'object',
     properties: {
-      categories: {
-        type: 'array',
-        items: { type: 'string' }
-      },
-      summary: { type: 'string' }
-    }
-  }
+      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
+// Stop a long prompt with a signal (when the caller holds the controller)
+const ac = new AbortController();
+const promptPromise = channel.prompt('Do a deep analysis', {
+  effort: 'RESEARCH',
+  signal: ac.signal,
+});
+ac.abort(); // asks the server to stop the in-flight interaction
+await promptPromise;
+```
 
-### Adding Users to a Space
+### Stopping interactions
 
-To add a user to a space, you need their user ID. Use `searchUser()` to find them by email:
+Use `signal` when the same call site cancels the prompt. When the Stop button
+lives elsewhere — a different component, after a reload, or a prompt another
+client started — stop by ID or stop the conversation's active interaction
+instead. Both are best-effort: the server halts the agent loop and closes the
+stream, but an LLM turn already in flight keeps generating server-side and is
+billed.
 
 ```typescript
-// Find the user by email
-const user = await client.searchUser('colleague@example.com');
-if (!user) {
-  throw new Error('User not found');
-}
+// Stop whatever is in flight on this channel's (default) conversation.
+// No-op returning false when nothing is running.
+await channel.stop();
 
-// 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
+// Stop a specific interaction by ID (e.g. from channel.activeLeafId or
+// the interactions list). Returns whether the server stopped it.
+await channel.stopInteraction(channel.activeLeafId!);
 
-These methods are available on `RoolSpace`:
+// Conversation handles stop their own in-flight interaction.
+const thread = channel.conversation('thread-42');
+await thread.stop();
+```
 
 | 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) |
+| --- | --- |
+| `stop(): Promise<boolean>` | Stop the in-flight interaction on the default conversation; `false` if none. |
+| `stopInteraction(id): Promise<boolean>` | Ask the server to stop a specific interaction by ID. |
+| `conversation.stop(): Promise<boolean>` | Stop a specific conversation's in-flight interaction. |
 
-### URL Sharing
+## Conversations
 
-Enable public URL access to allow anyone with the space URL to access it:
+Every channel has a default conversation. Use `channel.conversation(id)` for independent histories (for example, multiple chat threads). Conversations are represented as trees: interactions point at a `parentId`, and the SDK tracks an active leaf for each conversation.
 
 ```typescript
-const space = await client.openSpace('space-id');
+await channel.prompt('Hello'); // default conversation
 
-// Allow anyone with the URL to view
-await space.setLinkAccess('viewer');
-
-// Allow anyone with the URL to edit
-await space.setLinkAccess('editor');
+const thread = channel.conversation('thread-42');
+await thread.prompt('Hello from another thread');
+await thread.setSystemInstruction('Answer in haiku');
 
-// Disable URL access (default)
-await space.setLinkAccess('none');
+const branch = thread.getInteractions(); // active branch, root → leaf
+const tree = thread.getTree();           // full interaction tree
 
-// Check current setting
-console.log(space.linkAccess); // 'none' | 'viewer' | 'editor'
+if (thread.activeLeafId) {
+  thread.setActiveLeaf(thread.activeLeafId);
+}
 ```
 
-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/property | Description |
+| --- | --- |
+| `channel.conversation(id): ConversationHandle` | Get a conversation-scoped handle. |
+| `getInteractions(): Interaction[]` | Active branch as a flat list. |
+| `getTree(): Record<string, Interaction>` | Full interaction tree. |
+| `activeLeafId` | Current branch tip. |
+| `setActiveLeaf(id): void` | Switch branches. |
+| `getSystemInstruction()` / `setSystemInstruction(value)` | Manage conversation system instruction. Pass `null` to clear. |
+| `getConversations(): ConversationInfo[]` | List channel conversations (on `RoolChannel`). |
+| `deleteConversation(id): Promise<void>` | Delete a non-active conversation. |
+| `renameConversation(name): Promise<void>` | Rename the current/default conversation (on `RoolChannel`). |
+| `conversation.rename(name): Promise<void>` | Rename a specific conversation handle. |
 
-| 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, photoUrl, 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) |
+`ConversationHandle` also supports conversation-scoped `putObject`, `patchObject`, `moveObject`, `deleteObjects`, `prompt`, `stop`, collection-schema methods, and `setMetadata`.
 
-### Real-time Collaboration
+## Schema and Metadata
 
-When multiple users have a space open, changes sync in real-time. The `source` field in events tells you who made the change:
+Collections define the schema visible to the AI agent. Hidden body fields whose names start with `_` are useful for app/UI state that should not be considered by AI.
 
 ```typescript
-channel.on('objectUpdated', ({ objectId, object, source }) => {
-  if (source === 'remote_user') {
-    // Another user made this change
-    showCollaboratorActivity(object);
-  }
+await channel.createCollection('article', {
+  schemaOrgType: 'Article',
+  fields: [
+    { name: 'title', type: { kind: 'string' } },
+    { name: 'status', type: { kind: 'enum', values: ['draft', 'published'] } },
+    { name: 'tags', type: { kind: 'array', inner: { kind: 'string' } } },
+    { name: 'author', type: { kind: 'ref' } },
+  ],
 });
-```
-
-See [Real-time Sync](#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:
-
-```typescript
-// Default — errors only
-const client = new RoolClient();
+const schema = channel.getSchema();
 
-// Log everything to console
-const client = new RoolClient({ logger: console });
+await channel.alterCollection('article', [
+  { name: 'title', type: { kind: 'string' } },
+  { name: 'status', type: { kind: 'string' } },
+]);
 
-// Bring your own logger (pino, winston, etc.)
-const client = new RoolClient({
-  logger: myLogger // any object with { debug, info, warn, error }
-});
+channel.setMetadata('viewport', { x: 0, y: 0, zoom: 1 });
+const viewport = channel.getMetadata('viewport');
 ```
 
-### Space & Channel Lifecycle
-
 | Method | Description |
-|--------|-------------|
-| `listSpaces(): Promise<RoolSpaceInfo[]>` | List available spaces |
-| `openSpace(spaceId): Promise<RoolSpace>` | Open a space with live SSE subscription. Caches and reuses open spaces. Call `space.openChannel(channelId)` to get a channel. |
-| `createSpace(name): Promise<RoolSpace>` | Create a new space, returns live handle with SSE subscription |
-| `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
+| --- | --- |
+| `getSchema(): SpaceSchema` | Get collection definitions. |
+| `createCollection(name, fieldsOrDef, options?): Promise<CollectionDef>` | Create a collection. |
+| `alterCollection(name, fieldsOrDef, options?): Promise<CollectionDef>` | Replace a collection definition. |
+| `dropCollection(name): Promise<void>` | Remove a collection and its object directory. |
+| `setMetadata(key, value): void` | Set space metadata (fire-and-forget sync). |
+| `getMetadata(key): unknown` | Read metadata from local cache. |
+| `getAllMetadata(): Record<string, unknown>` | Read all metadata from local cache. |
 
-Manage channels on the `RoolSpace` handle:
-
-| Method | Description |
-|--------|-------------|
-| `space.channels: ChannelInfo[]` | Live channel list (auto-updates via SSE) |
-| `space.getChannels(): ChannelInfo[]` | List channels (deprecated — use `space.channels` instead) |
-| `space.renameChannel(channelId, name): Promise<void>` | Rename a channel |
-| `space.deleteChannel(channelId): Promise<void>` | Delete a channel and its interaction history |
-| `channel.rename(name): Promise<void>` | Rename the current open channel |
+Field kinds: `string`, `number`, `boolean`, `ref`, `enum`, `literal`, `array`, and `maybe`.
 
-### User Storage
+## Undo/Redo
 
-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) |
+Undo/redo uses checkpoints for the current channel ID. A checkpoint captures space state; call `checkpoint()` before a user action you want to make undoable.
 
 ```typescript
-// 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);
+await channel.checkpoint('Delete article');
+await channel.deleteObjects(['/space/article/welcome.json']);
 
-// 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);
-});
+if (await channel.canUndo()) {
+  await channel.undo();
+}
 ```
 
-### Extensions
-
-Manage and publish extensions. See [`@rool-dev/extension`](/extension/) for building extensions.
-
-There are two distinct domains: your **personal library** (extensions you've created or installed) and the **published extensions** (extensions discoverable by all users). Each has its own return type.
-
-#### Your Library (`ExtensionInfo`)
-
-Manage extensions you own. Each `ExtensionInfo` includes `published` (whether it's listed in the marketplace) and `marketplaceExtensionId` (non-null if you installed it from someone else's listing, null if you authored it).
-
-| Method | Description |
-|--------|-------------|
-| `uploadExtension(extensionId, options): Promise<ExtensionInfo>` | Upload or update an extension (`options.bundle`: zip with `index.html` and `manifest.json`) |
-| `listExtensions(): Promise<ExtensionInfo[]>` | List your extensions |
-| `getExtensionInfo(extensionId): Promise<ExtensionInfo \| null>` | Get info for a specific extension |
-| `deleteExtension(extensionId): Promise<void>` | Delete an extension permanently (removes files and DB row) |
-
-#### Marketplace (`PublishedExtensionInfo`)
-
-Discover and install extensions published by other users.
-
-| Method | Description |
-|--------|-------------|
-| `findExtensions(options?): Promise<PublishedExtensionInfo[]>` | Search the marketplace. 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. If you own it, wires it directly. If it's a marketplace extension, copies and builds a new extension in your library. Returns the channel ID. |
-| `publishToPublic(extensionId): Promise<void>` | Publish one of your extensions to the marketplace |
-| `unpublishFromPublic(extensionId): Promise<void>` | Remove from the marketplace (keeps the extension in your library) |
-
-### Utilities
-
 | Method | Description |
-|--------|-------------|
-| `RoolClient.generateId(): string` | Generate 6-char alphanumeric ID (static) |
-| `destroy(): void` | Clean up resources |
+| --- | --- |
+| `checkpoint(label?): Promise<string>` | Save current space state. |
+| `canUndo(): Promise<boolean>` | Check whether undo is available. |
+| `canRedo(): Promise<boolean>` | Check whether redo is available. |
+| `undo(): Promise<boolean>` | Restore the latest checkpoint. |
+| `redo(): Promise<boolean>` | Reapply undone work. |
+| `clearHistory(): Promise<void>` | Clear checkpoint history. |
 
-### Client Events
+Undo/redo availability and history are scoped to the channel handle (`channel.channelId`).
 
-```typescript
-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('channelUpdated', (spaceId: string, channel: ChannelInfo) => 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)
-```
+## File Storage and WebDAV
 
-Channel events on the client (`channelCreated`, `channelUpdated`, `channelDeleted`) are pass-throughs from space events for backwards compatibility. Prefer listening on the space handle directly for new code.
+Every space has authenticated WebDAV storage. WebDAV methods take SDK machine paths such as `/space/...`, `/rool-drive/...`, or `/` for the root collection.
 
-**Space list management pattern:**
 ```typescript
-const spaces = new Map<string, RoolSpaceInfo>();
+const webdav = space.webdav;
 
-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 });
+await webdav.mkcol('/rool-drive/docs');
+await webdav.put('/rool-drive/docs/readme.md', '# Hello', {
+  contentType: 'text/markdown',
+  ifNoneMatch: '*',
 });
-```
-
-## RoolSpace API
-
-A space handle with a live SSE subscription. Extends `EventEmitter`. Manages user access, link sharing, channels, and export. The `channels` property auto-updates via SSE, and channel lifecycle events fire in real-time.
 
-`openSpace()` caches and reuses open spaces — calling it twice with the same ID returns the same instance. Call `close()` when done to stop the subscription and close all open channels.
-
-### Properties
+const listing = await webdav.propfind('/rool-drive/docs', {
+  depth: '1',
+  props: ['displayname', 'getcontentlength', 'getcontenttype', 'getetag'],
+});
 
-| 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 |
-| `channels: ChannelInfo[]` | Live channel list (auto-updates via SSE) |
+const response = await webdav.get('/rool-drive/docs/readme.md');
+console.log(await response.text());
 
-### Methods
+const file = await space.fetchPath('/rool-drive/docs/readme.md');
+console.log(file.headers.get('Content-Type'));
 
-| Method | Description |
-|--------|-------------|
-| `openChannel(channelId): Promise<RoolChannel>` | Open a channel on this space |
-| `close(): void` | Stop SSE subscription and close all open channels |
-| `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 (deprecated — use `channels` property instead) |
-| `renameChannel(channelId, name): Promise<void>` | Rename a channel |
-| `deleteChannel(channelId): Promise<void>` | Delete a channel |
-| `exportArchive(): Promise<Blob>` | Export space as zip archive |
-| `refresh(): Promise<void>` | Refresh space data from server |
-
-### Space Events
-
-```typescript
-space.on('channelCreated', (channel: ChannelInfo) => void)   // New channel added
-space.on('channelUpdated', (channel: ChannelInfo) => void)   // Channel metadata changed (name, extension, manifest)
-space.on('channelDeleted', (channelId: string) => void)      // Channel removed
-space.on('connectionStateChanged', (state: 'connected' | 'disconnected' | 'reconnecting') => void)
+const usage = await space.getStorageUsage();
+console.log(usage.usedBytes, usage.availableBytes, usage.limitBytes);
 ```
 
-## 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 |
-| `extensionId: string \| null` | ID of the installed extension, or null if this is a plain channel |
-| `manifest: ExtensionManifest \| null` | Extension manifest snapshot (name, icon, collections, etc.), or null |
-
-### 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](#conversations)) |
-
-### Object Operations
-
-Objects are plain key/value records. `id` and `type` are reserved; everything else is application-defined. References between objects are data fields whose values are object IDs. Every object must include a `type` field whose value names a collection in the schema (see [Collection Schema](#collection-schema)) — that binds the object to that collection. Before introducing a new kind of object, create the matching collection.
-
-| 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, and the channel/conversation/interaction where the last write happened), 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). Must include `type` naming an existing collection. 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
+### Real-time file sync
 
-| Option | Description |
-|--------|-------------|
-| `data` | Fields to add or update. Pass `null`/`undefined` to delete a field. Use `{{placeholder}}` for AI-generated content. Setting a new `type` retypes the object — the merged result must conform to the new collection. 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 (matches objects whose `type` field equals the name), 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. Returns objects whose `type` field equals the given name. |
-| `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:**
+Object and file changes are announced at the space level. Use WebDAV `syncCollection()` to reconcile changes.
 
 ```typescript
-// Filter by collection (no AI, no credits)
-const { objects } = await channel.findObjects({
-  collection: 'article'
-});
+let token: string | null = null;
 
-// 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' }
-});
+async function syncFiles() {
+  const result = await space.webdav.syncCollection('/', {
+    token,
+    level: 'infinite',
+    props: ['displayname', 'getetag', 'getlastmodified', 'resourcetype'],
+  });
+  token = result.token;
+  updateFileTree(result.responses);
+}
 
-// Pure natural language query (AI interprets)
-const { objects, message } = await channel.findObjects({
-  prompt: 'articles about space exploration published this year'
+space.on('filesChanged', syncFiles);
+space.on('filesReset', () => {
+  token = null;
+  void syncFiles();
 });
 
-// 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
-});
+await syncFiles();
 ```
 
-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](#checkpoints--undoredo) 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.
+| --- | --- |
+| `webdav.href(path)` / `webdav.url(path)` | Return WebDAV href/URL for an absolute SDK path. |
+| `webdav.options(path)` | Send `OPTIONS`. |
+| `webdav.propfind(path, options)` | Read properties/list collections. `depth` is required. |
+| `webdav.syncCollection(path, options)` | WebDAV `REPORT sync-collection`; returns changed responses and next token. |
+| `webdav.get(path, options?)` / `webdav.head(path)` | Read a file; `get` supports byte ranges. |
+| `webdav.put(path, body, options?)` | Write a file/object at an exact path. Parent collection must exist. |
+| `webdav.mkcol(path)` | Create one collection. |
+| `webdav.copy(source, destination, options?)` | Copy a file or collection. |
+| `webdav.move(source, destination, options?)` | Move a file or collection. |
+| `webdav.delete(path, options?)` | Delete a file or collection. |
+| `webdav.lock(path, options)` / `refreshLock(path, token)` / `unlock(token)` | WebDAV write locks. |
+| `webdav.request(method, path, init?)` | Raw authenticated WebDAV request. |
+| `space.fetchPath(path, options?)` | Fetch a `/rool-drive/...` file path or `rool-machine:` file URI. |
+| `space.getStorageUsage()` / `webdav.getStorageUsage()` | Storage quota usage. |
+
+High-level WebDAV methods that validate response status throw `WebDAVError` with `status`, `statusText`, and `body`; raw `request()` and `options()` return `Response`.
 
-| 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 |
+## Collaboration
 
 ```typescript
-// Upload an image
-const url = await channel.uploadMedia(file);
-await channel.createObject({ data: { type: 'photo', title: 'Photo', image: url } });
-
-// Or let AI generate one using a placeholder
-await channel.createObject({
-  data: { type: 'photo', 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);
+const user = await client.searchUser('colleague@example.com');
+if (user) {
+  await space.addUser(user.id, 'editor');
 }
-```
-
-### Proxied Fetch
-
-Fetch external URLs via the server, bypassing CORS restrictions. Requires editor role or above. Private/internal IP ranges are blocked (SSRF protection).
-
-| Method | Description |
-|--------|-------------|
-| `fetch(url, init?): Promise<Response>` | Fetch a URL via the server proxy. `init` accepts `method`, `headers`, and `body`. |
 
-```typescript
-// GET request
-const response = await channel.fetch('https://api.example.com/data');
-const data = await response.json();
-
-// POST with headers and body
-const response = await channel.fetch('https://api.example.com/submit', {
-  method: 'POST',
-  headers: { 'Content-Type': 'application/json' },
-  body: { key: 'value' },
-});
+await space.setLinkAccess('viewer'); // 'none' | 'viewer' | 'editor'
 ```
 
-### Collection Schema
+Roles:
 
-Collections are the types you use to group objects in a space. Every object must belong to a collection: the object's `data.type` field names the collection it belongs to, and the server validates the object's fields against that collection's definition. Renaming a collection cascades to the `type` field of every object bound to it; dropping a collection is blocked while any object is still bound to it.
+| Role | Capabilities |
+| --- | --- |
+| `owner` | Full control. |
+| `admin` | Editor capabilities plus user/link management. |
+| `editor` | Create, modify, move, and delete objects/files. |
+| `viewer` | Read-only access. |
 
-Collections make up the schema and are stored in the space data, syncing in real time. The schema is visible to the AI agent so it knows which collections exist and what fields they contain, producing more consistent objects.
+## RoolClient API
 
+### Constructor config
 
 ```typescript
-// 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');
+const client = new RoolClient({
+  apiUrl: 'https://api.rool.dev',
+  authUrl: 'https://rool.dev/auth',
+  graphqlUrl: 'https://api.rool.dev/graphql',
+  logger: console,
+});
 ```
 
-| 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' } }` |
+`apiUrl` defaults to `https://api.rool.dev`; `authUrl` is derived by stripping the `api.` hostname prefix unless provided. `baseUrl` is still accepted as a deprecated alias for `apiUrl`. Pass `authProvider` for Node.js, Electron, or custom auth flows.
+
+| Method/property | Description |
+| --- | --- |
+| `currentUser: CurrentUser | null` | Cached user profile from initialization/fetch. |
+| `getCurrentUser(): Promise<CurrentUser>` | Fetch current user. |
+| `updateCurrentUser(input): Promise<CurrentUser>` | Update `name`, `slug`, or `marketingOptIn`. |
+| `deleteCurrentUser(): Promise<void>` | Mark account for deletion and log out. |
+| `searchUser(email): Promise<UserResult | null>` | Exact email lookup. |
+| `listSpaces(): Promise<RoolSpaceInfo[]>` | List accessible spaces. |
+| `openSpace(id): Promise<RoolSpace>` | Open/cached live space handle. |
+| `createSpace(name): Promise<RoolSpace>` | Create and open a space. |
+| `duplicateSpace(sourceId, name): Promise<RoolSpace>` | Duplicate a space. |
+| `deleteSpace(id): Promise<void>` | Permanently delete a space. |
+| `importArchive(name, archive): Promise<RoolSpace>` | Import a zip archive as a new space. |
+| `getUserStorage<T>(key): T | undefined` | Sync read from user-storage cache. |
+| `setUserStorage(key, value): void` | Update user storage; `null`/`undefined` deletes. |
+| `getAllUserStorage(): Record<string, unknown>` | Copy all cached user storage. |
+| `reportEvent(event, url?): void` | Fire-and-forget telemetry event. |
+| `destroy(): void` | Close subscriptions, spaces, auth resources, and listeners. |
+| `generateId(): string` | Generate a 6-character alphanumeric ID. |
+
+### Client events
+
+```typescript
+client.on('authStateChanged', (authenticated) => void 0);
+client.on('spaceAdded', (space) => void 0);
+client.on('spaceRemoved', (spaceId) => void 0);
+client.on('spaceRenamed', (spaceId, newName) => void 0);
+client.on('channelCreated', (spaceId, channel) => void 0);
+client.on('channelUpdated', (spaceId, channel) => void 0);
+client.on('channelDeleted', (spaceId, channelId) => void 0);
+client.on('userStorageChanged', ({ key, value, source }) => void 0);
+client.on('connectionStateChanged', (state) => void 0);
+client.on('error', (error, context) => void 0);
+```
 
-### Import/Export
+## RoolSpace API
 
-Export and import space data as zip archives for backup, portability, or migration:
+Properties: `id`, `name`, `role`, `linkAccess`, `memberCount`, `channels`, `route`, `webdav`.
 
 | 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:**
-```typescript
-const space = await client.openSpace('space-id');
-const archive = await space.exportArchive();
-// Save as .zip file
-const url = URL.createObjectURL(archive);
+| --- | --- |
+| `openChannel(channelId): Promise<RoolChannel>` | Open/create a channel. |
+| `close(): void` | Stop subscription and close open channels. |
+| `rename(newName): Promise<void>` | Rename the space. |
+| `delete(): Promise<void>` | Permanently delete the space. |
+| `listUsers(): Promise<SpaceMember[]>` | List collaborators. |
+| `addUser(userId, role): Promise<void>` | Add collaborator. |
+| `removeUser(userId): Promise<void>` | Remove collaborator. |
+| `setLinkAccess(linkAccess): Promise<void>` | Set URL sharing level. |
+| `renameChannel(channelId, name): Promise<void>` | Rename a channel. |
+| `deleteChannel(channelId): Promise<void>` | Delete a channel and history. |
+| `exportArchive(): Promise<Blob>` | Export a space archive. |
+| `refresh(): Promise<void>` | Refresh cached space data. |
+| `fetchPath(path, options?): Promise<Response>` | Fetch a `/rool-drive/...` file. |
+| `getStorageUsage(): Promise<SpaceFileStorageUsage>` | File-storage quota usage. |
+
+Events:
+
+```typescript
+space.on('channelCreated', (channel) => void 0);
+space.on('channelUpdated', (channel) => void 0);
+space.on('channelDeleted', (channelId) => void 0);
+space.on('filesChanged', ({ spaceId, source, timestamp }) => void 0);
+space.on('filesReset', ({ spaceId, source, timestamp }) => void 0);
+space.on('connectionStateChanged', (state) => void 0);
 ```
 
-**Import:**
-```typescript
-const space = await client.importArchive('Imported Data', archiveBlob);
-const channel = await space.openChannel('main');
-```
+## RoolChannel API
 
-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.
+Properties: `id` (space ID), `name` (space name), `role`, `linkAccess`, `userId`, `channelId`, `channelName`, `conversationId`, `isReadOnly`, `activeLeafId`.
 
-### Channel Events
+| Area | Methods |
+| --- | --- |
+| Lifecycle | `close()`, `rename(name)`, `conversation(id)` |
+| Objects | `getObject`, `getObjects`, `stat`, `putObject`, `patchObject`, `moveObject`, `deleteObjects` |
+| Schema | `getSchema`, `createCollection`, `alterCollection`, `dropCollection` |
+| Metadata | `setMetadata`, `getMetadata`, `getAllMetadata` |
+| Conversations | `getInteractions`, `getTree`, `setActiveLeaf`, `getConversations`, `deleteConversation`, `getSystemInstruction`, `setSystemInstruction`, `renameConversation` |
+| AI | `prompt`, `stop`, `stopInteraction` |
+| Undo/redo | `checkpoint`, `canUndo`, `canRedo`, `undo`, `redo`, `clearHistory` |
+| Utilities | `fetch(url, init?)` server-side proxied fetch |
 
-Semantic events describe what changed. Events fire for both local changes and remote changes.
+Channel events:
 
 ```typescript
-// 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)
+channel.on('metadataUpdated', ({ metadata, source }) => void 0);
+channel.on('schemaUpdated', ({ schema, source }) => void 0);
+channel.on('channelUpdated', ({ channelId, source }) => void 0);
+channel.on('conversationUpdated', ({ conversationId, channelId, source }) => void 0);
+channel.on('reset', ({ source }) => void 0);
+channel.on('syncError', (error) => void 0);
 ```
 
-### Error Handling
+`channel.fetch(url, init?)` proxies external HTTP requests through the server to bypass browser CORS.
 
-AI operations may fail due to rate limiting or other transient errors. Check `error.message` for user-friendly error text:
+## Import/Export
 
 ```typescript
-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);
-  }
-}
+const archive = await space.exportArchive();
+const imported = await client.importArchive('Imported Data', archive);
 ```
 
-## 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](#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.
+Archives include objects, metadata, channels/conversations, and file storage.
 
 ## Data Types
 
-### Schema Types
-
 ```typescript
-// Allowed field types
 type FieldType =
   | { kind: 'string' }
   | { kind: 'number' }
@@ -1108,141 +643,66 @@ interface FieldDef {
 
 interface CollectionDef {
   fields: FieldDef[];
+  schemaOrgType?: string;
 }
 
-// Full schema — collection names to definitions
 type SpaceSchema = Record<string, CollectionDef>;
-```
-
-### Object Data
 
-```typescript
-// 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;
+  path: string;
+  body: Record<string, unknown>;
+}
+
+interface GetObjectsResult {
+  objects: RoolObject[];
+  missing: string[];
 }
 
-// Object stat - audit information returned by channel.stat()
 interface RoolObjectStat {
+  path: string;
   modifiedAt: number;
   modifiedBy: string;
   modifiedByName: string | null;
-  modifiedInChannel: string;                    // Channel ID where the last modification happened
-  modifiedInConversation: string | null;        // Conversation ID, or null if not conversation-scoped
-  modifiedInInteraction: string | null;         // Interaction ID, or null for ephemeral or non-AI writes
-}
-```
-
-### Channels and Conversations
-
-```typescript
-// 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)
-  extensionId?: string;         // ID of installed extension (user_extensions.extension_id)
-  manifest?: ExtensionManifest; // Extension manifest snapshot (set when extension is wired)
-  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
-  extensionId: string | null;   // ID of installed extension, or null
-  manifest: ExtensionManifest | null;  // Extension manifest snapshot, or null
+  modifiedInChannel: string;
+  modifiedInConversation: string | null;
+  modifiedInInteraction: string | 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.
+type PromptAttachment =
+  | File
+  | Blob
+  | { data: string; contentType: string; filename?: string }
+  | string;
 
-### Interaction Types
+type PromptEffort = 'QUICK' | 'STANDARD' | 'REASONING' | 'RESEARCH';
 
-```typescript
-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)
+interface PromptOptions {
+  responseSchema?: Record<string, unknown>;
+  effort?: PromptEffort;
+  parentInteractionId?: string | null;
+  ephemeral?: boolean;
+  readOnly?: boolean;
+  attachments?: PromptAttachment[];
+  signal?: AbortSignal;
+  eventName?: string;
 }
 
 type InteractionStatus = 'pending' | 'streaming' | 'done' | 'error';
 
 interface Interaction {
-  id: string;                    // Unique ID for this interaction
-  parentId: string | null;       // Parent in conversation tree (null = root)
+  id: string;
+  parentId: string | null;
   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
-
-```typescript
-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; photoUrl: string | null; }
-interface UserResult { id: string; email: string; name: string | null; photoUrl: string | null; }
-interface CurrentUser { id: string; email: string; name: string | null; photoUrl: 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
-
-```typescript
-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)
+  userId: string;
+  userName?: string | null;
+  operation: 'prompt' | 'putObject' | 'patchObject' | 'moveObject' | 'deleteObjects' | 'deletePaths' | string;
+  input: string;
+  output: string | null;
+  status: InteractionStatus;
+  ai: boolean;
+  modifiedObjectPaths: string[];
+  toolCalls: ToolCall[];
+  attachments?: string[];
 }
 ```