Rw Integrate Documents

Constraints

• Each Avatar supports up to 50,000 tokens of knowledge
• Supported formats: plain text and Markdown
• More formats planned for future releases

Flow

The flow is: Create a Document → Link it to an Avatar

Step 1: Create a Document

Node.js

import RunwayML from '@runwayml/sdk';

const client = new RunwayML();

const document = await client.documents.create({
  name: 'Product FAQ',
  content: `# Product FAQ

## What is your return policy?
We offer a 30-day return policy for all unused items in original packaging.

## How do I track my order?
Log in to your account and visit the Orders page. You'll find tracking information for all shipped orders.

## Do you offer international shipping?
Yes, we ship to over 50 countries. Shipping costs and delivery times vary by destination.`,
});

console.log('Document created:', document.id);

Python

from runwayml import RunwayML

client = RunwayML()

document = client.documents.create(
    name='Product FAQ',
    content="""# Product FAQ

## What is your return policy?
We offer a 30-day return policy for all unused items in original packaging.

## How do I track my order?
Log in to your account and visit the Orders page.

## Do you offer international shipping?
Yes, we ship to over 50 countries.""",
)

print('Document created:', document.id)

Step 2: Link the Document to an Avatar

Update your Avatar to attach the document. This replaces any existing document attachments — pass all document IDs you want linked.

Node.js

await client.avatars.update(avatarId, {
  documentIds: [document.id],
});

Python

client.avatars.update(
    avatar_id,
    document_ids=[document.id],
)

Multiple Documents

You can link multiple documents to a single avatar (total must stay under 50,000 tokens):

const faq = await client.documents.create({
  name: 'FAQ',
  content: '...',
});

const policies = await client.documents.create({
  name: 'Company Policies',
  content: '...',
});

await client.avatars.update(avatarId, {
  documentIds: [faq.id, policies.id],
});

Step 3: Start a Session

Once documents are linked, the Avatar automatically has access to the knowledge during conversations. Start a session as usual — no additional configuration needed:

const session = await client.realtimeSessions.create({
  model: 'gwm1_avatars',
  avatar: {
    type: 'custom',
    avatarId: avatarId,
  },
});

session = client.realtime_sessions.create(
    model='gwm1_avatars',
    avatar={
        'type': 'custom',
        'avatar_id': avatar_id,
    },
)

See +rw-integrate-characters for the full session creation, polling, and WebRTC flow.

Integration Patterns

Load Documents from Files

Read content from local files and create documents:

import fs from 'fs';
import RunwayML from '@runwayml/sdk';

const client = new RunwayML();

// Read a local markdown file
const content = fs.readFileSync('./knowledge/product-faq.md', 'utf-8');

const document = await client.documents.create({
  name: 'Product FAQ',
  content,
});

from pathlib import Path
from runwayml import RunwayML

client = RunwayML()

content = Path('./knowledge/product-faq.md').read_text()

document = client.documents.create(
    name='Product FAQ',
    content=content,
)

Dynamic Knowledge Updates

Update documents programmatically — useful for syncing with a CMS or database:

// Example: API endpoint to update character knowledge
app.post('/api/avatar/update-knowledge', async (req, res) => {
  const { avatarId, documents } = req.body;

  // Create new documents
  const docIds = [];
  for (const doc of documents) {
    const created = await client.documents.create({
      name: doc.name,
      content: doc.content,
    });
    docIds.push(created.id);
  }

  // Link all documents to the avatar (replaces existing)
  await client.avatars.update(avatarId, {
    documentIds: docIds,
  });

  res.json({ success: true, documentCount: docIds.length });
});

Tips

• Use Markdown for structured content — headings help the Avatar navigate the knowledge.
• Be concise — stay well under the 50,000 token limit for best retrieval quality.
• Organize by topic — multiple focused documents work better than one giant document.
• Update regularly — keep knowledge current by re-creating documents when content changes.
• Documents can also be managed via the Developer Portal UI.