Voice Sessions Now Know Who They Are. Here's What Changed.

OpenClaw 2026.5.21 made a change to how voice sessions work that I initially glossed over but think is actually significant: realtime voice sessions now include bounded IDENTITY.md, USER.md, and SOUL.md profile context in the session instructions by default.

This sounds small. It isn't.

What It Means in Practice

Before this change, a voice session started fresh. It had no context about who it was, who it was talking to, or what its persona was supposed to be. You could tell it in the conversation, but it wasn't baked into the session instructions.

After this change, voice sessions bootstrap with your IDENTITY.md (who you are), USER.md (who you're helping), and SOUL.md (how you behave) as bounded context. It's included in the session instructions by default.

If you've defined your persona in these files — which you should have — your voice sessions now actually know who they are.

How to Disable It If You Need To

If you have a use case where you specifically want a voice session to NOT have access to the profile context (maybe for privacy reasons, maybe for a specific automation), there's a config knob:

{
  "voice": {
    "realtime": {
      "bootstrapContextFiles": []
    }
  }
}

Setting bootstrapContextFiles to an empty array disables the profile context injection. This is useful for anonymous voice automation where the agent shouldn't have access to user profile data.

The Security Implication

This is a security-positive change. When voice sessions have bounded, explicit context, they don't have to guess or infer who they are. The bounded context means they can't access profile data outside what's explicitly included in the bootstrap.

If you've ever had a voice session that seemed confused about who it was or what it was supposed to do, this is likely why — it was starting without profile context and inferring from conversation, which is unreliable.

The Voice Session Follow Feature

The same release added Discord voice sessions that can follow configured users into voice channels. This is a separate feature, but related: it means you can configure a Discord user, and when that user joins a voice channel, the OpenClaw agent can join with them.

Use cases:

• Voice agent that sits in your personal Discord voice channel and responds to voice commands
• Team voice channel where the agent is present when specific people are present
• Automated voice summaries after meetings in voice channels

This requires explicit opt-in per channel — the agent won't just join any voice channel.

The Config to Enable Voice Session Context

If you're running OpenClaw 2026.5.21 or later and want voice sessions to have profile context:

1. Make sure your IDENTITY.md, USER.md, and SOUL.md are populated with accurate information
2. Upgrade to 2026.5.21 (or later)
3. No additional config needed — it's on by default

If you want to verify it's working, start a voice session and ask the agent "who are you and who are you helping?" If it answers correctly, the profile context is working.

---

Voice sessions with profile context: OpenClaw 2026.5.21. Discord voice follow: same release, requires per-channel opt-in.