feat: Add Moss plugin for sub 10ms semantic search integration

[yatharthk2]

Summary

This PR adds a new plugin for Moss semantic search integration into the LiveKit Agents ecosystem. The plugin provides a thin wrapper around the Moss SDK with LiveKit-friendly defaults, environment-based configuration, and index caching.

Quick Feature List of Moss :

1. Sub-10ms semantic search — Retrieval from device memory for real-time voice agent responses
2. Offline-first with cloud sync — Local searches work offline; optional cloud sync for persistence
3. Multi-index management — Create, query, and manage multiple isolated search spaces

What's Included

• New plugin package: livekit-plugins-moss
• MossClient: Async wrapper for semantic search operations
• Full index lifecycle support: Create, load, query, and manage indexes
• Document management: Add, retrieve, and delete documents
• Environment-based configuration: Uses MOSS_PROJECT_ID and MOSS_PROJECT_ID environment variables
• Example implementation: Full lifecycle demo in examples/dev/MossLifecycle.py

Usage Example

from livekit.plugins.moss import DocumentInfo, MossClient

client = MossClient()
await client.create_index(
    "demo",
    [DocumentInfo(id="doc1", text="Semantic search is fast")],
    model_id="moss-minilm",
)
result = await client.query("demo", "fast search", top_k=5)

Dependencies

• inferedge-moss>=1.0.0b2

Related

• Example implementation: examples/dev/MossLifecycle.py
• Moss documentation: Moss Docs