VML

xAI gave Grok Voice. We gave Grok memory.

What it is

A drop-in Python SDK that adds persistent memory to LiveKit + xAI Grok voice agents. One import, one binding — the agent stops forgetting between sessions.

Install: pip install grok-faf-voice

Why it exists

Voice AI agents start every conversation from zero. Names, preferences, context — gone. Every. Time.

VML is the memory layer that fixes that. Persistent state lives outside the agent, etched to a private namespace, recalled on demand. Memory survives sessions, devices, and model switches. The agent gets continuity it never had.

Receipts

  • On PyPIgrok-faf-voice v0.2.2
  • LiveKit-enabled — cascaded pipeline, drop-in
  • Sessionless MCP transport — talks to mcpaas.live via X-MCP-Mode: flexi
  • FAFm-powered — memories stored as application/vnd.fafm+yaml, IANA-registered May 13, 2026

Try it

The multi-model voice frontend uses VML for its memory layer. Speak to Leo, switch models, your facts persist.

Talk to FAFA →

FAF Agent is called FAFA (rhymes with RAFA).