Using MacOs voice-service instead of ElevenLabs

[Xantoshi-BTC]

Hi all,

Love the idea and promise of PAI, very excited about it!

Curious to understand if any of you have been able to install PAI using the MacOs voice instead of ElevenLabs?
The installation itself runs fine it seems but I noticed the config files still have a lot of references to /.claude

Per update

📅 October 2, 2025 - Voice System Update

• 🎙️ Zero-Cost Voice System: Migrated from ElevenLabs API to macOS native Premium/Enhanced voices

Instead of manually updating every file to "/PAI/PAI_DIRECTORY/voice-server/" is there an easy way to replace or re-configure the files so that I don't need to do this manually?

examples include:
VOICE_SERVER_DIR="$HOME/.claude/voice-server"
"PAI_DIR": "/Users/daniel/.claude"

Thanks in advance!

Xantoshi

[smolcompute]

I'm planning on trying to use MacOS voices too for privacy reasons. It seems @danielmiessler first changed the PAI repo to migrate from ElevenLabs to MacOS, then a few days later reverted back to ElevenLabs, but I do not know why. Did it not work well ?

[Xantoshi-BTC]

Yeah I am also curious about this. @danielmiessler besides the questions why you changed it back to ElevenLabs, can we help you make this setup more generic? e.g. for using Codex? --> I have privacy concerns so would like to use my own models (local Ollama or MapleAI)

PS:

> grep -rn '.' -e '~/.claude'

> > ./PAI_DIRECTORY/documentation/voice-system.md:10:1. **Voice Server** (`~/.claude/voice-server/server.ts`)
> > ./PAI_DIRECTORY/documentation/voice-system.md:17:2. **Stop Hook** (`~/.claude/hooks/stop-hook.ts`)
> > ./PAI_DIRECTORY/documentation/voice-system.md:24:3. **Agent Configurations** (`~/.claude/agents/*.md`)
> > ./PAI_DIRECTORY/documentation/voice-system.md:90:**Location:** `~/.claude/voice-server/voices.json`
> > ./PAI_DIRECTORY/documentation/voice-system.md:197:Each agent file (`~/.claude/agents/*.md`) includes voice configuration:
> > ./PAI_DIRECTORY/documentation/voice-system.md:327:   grep "const VOICES" ~/.claude/hooks/stop-hook.ts
> > ./PAI_DIRECTORY/documentation/voice-system.md:332:   grep "voiceId:" ~/.claude/agents/*.md
> > ./PAI_DIRECTORY/documentation/VOICE-SETUP-GUIDE.md:212:cd ~/.claude/voice-server
> > ./PAI_DIRECTORY/documentation/VOICE-SETUP-GUIDE.md:330:cd ~/.claude/voice-server
> > ./PAI_DIRECTORY/documentation/VOICE-SETUP-GUIDE.md:339:grep "const VOICES" ~/.claude/hooks/stop-hook.ts
> > ./PAI_DIRECTORY/documentation/VOICE-SETUP-GUIDE.md:342:grep "voiceId:" ~/.claude/agents/*.md
> > ./PAI_DIRECTORY/documentation/VOICE-SETUP-GUIDE.md:356:2. Update the voice mapping in `~/.claude/hooks/stop-hook.ts`:
> > ./PAI_DIRECTORY/documentation/VOICE-SETUP-GUIDE.md:438:cd ~/.claude/voice-server && bun server.ts &
> > ./PAI_DIRECTORY/documentation/VOICE-SETUP-GUIDE.md:448:- Voice server: `~/.claude/voice-server/server.ts`
> > ./PAI_DIRECTORY/documentation/VOICE-SETUP-GUIDE.md:449:- Stop hook: `~/.claude/hooks/stop-hook.ts`
> > ./PAI_DIRECTORY/documentation/VOICE-SETUP-GUIDE.md:450:- Agent configs: `~/.claude/agents/*.md`
> > ./PAI_DIRECTORY/documentation/VOICE-SETUP-GUIDE.md:451:- Documentation: `~/.claude/documentation/voice-system.md`
> > ./PAI_DIRECTORY/agents/architect.md:28:   - Read `~/.claude/context/CLAUDE.md` - The complete context system documentation
> > ./PAI_DIRECTORY/agents/designer.md:30:   - `read ~/.claude/context/CLAUDE.md` - The complete context system documentation
> > ./PAI_DIRECTORY/agents/designer.md:31:   - `read ~/.claude/context/tools/CLAUDE.md` - The tools you have available
> > ./PAI_DIRECTORY/agents/designer.md:83:Recall the Chrome DevTools sections in your already loaded context: `~/.claude/context/tools/CLAUDE.md`
> > ./PAI_DIRECTORY/agents/researcher.md:19:- `~/.claude/commands/web-research.md` - This is your PRIMARY AND ONLY research tool!!!
> > ./PAI_DIRECTORY/agents/pentester.md:26:   - Read `~/.claude/context/CLAUDE.md` - The complete context system documentation
> > ./PAI_DIRECTORY/agents/engineer.md:29:   - `read ~/.claude/context/CLAUDE.md` - The complete context system documentation
> > ./PAI_DIRECTORY/agents/engineer.md:30:   - `read ~/.claude/context/tools/CLAUDE.md` - The tools you have available
> > ./PAI_DIRECTORY/voice-server/macos-service/uninstall.sh:35:echo "  ~/.claude/voice-server/logs/"
> > ./PAI_DIRECTORY/voice-server/macos-service/validate-setup.sh:66:    "[ -f ~/.claude/voice-server/server.ts ]" \
> > ./PAI_DIRECTORY/voice-server/macos-service/validate-setup.sh:67:    "Server file missing at ~/.claude/voice-server/server.ts"
> > ./PAI_DIRECTORY/voice-server/macos-service/validate-setup.sh:139:                echo "   Check logs: tail -f ~/.claude/voice-server/logs/voice-server-error.log"
> > ./PAI_DIRECTORY/voice-server/menubar/pai-voice.5s.sh:27:    echo "Stop Server | bash='~/.claude/voice-server/stop.sh' terminal=false refresh=true"
> > ./PAI_DIRECTORY/voice-server/menubar/pai-voice.5s.sh:28:    echo "Restart Server | bash='~/.claude/voice-server/restart.sh' terminal=false refresh=true"
> > ./PAI_DIRECTORY/voice-server/menubar/pai-voice.5s.sh:35:    echo "Start Server | bash='~/.claude/voice-server/start.sh' terminal=false refresh=true"
> > ./PAI_DIRECTORY/voice-server/menubar/pai-voice.5s.sh:39:echo "Check Status | bash='~/.claude/voice-server/status.sh' terminal=true"
> > ./PAI_DIRECTORY/voice-server/menubar/pai-voice.5s.sh:44:echo "Open Voice Server Folder | bash='open ~/.claude/voice-server'"
> > ./PAI_DIRECTORY/voice-server/menubar/pai-voice.5s.sh:45:echo "Uninstall | bash='~/.claude/voice-server/uninstall.sh' terminal=true"
> > ./PAI_DIRECTORY/voice-server/QUICKSTART.md:134:tail -f ~/.claude/voice-server/logs/voice-server.log
> > ./PAI_DIRECTORY/voice-server/QUICKSTART.md:217:2. Review server logs: `tail -f ~/.claude/voice-server/logs/voice-server.log`
> > ./PAI_DIRECTORY/voice-server/README.md:228:grep "VOICE_CONFIG" ~/.claude/hooks/stop-hook.ts
> > ./PAI_DIRECTORY/voice-server/README.md:247:~/.claude/hooks/
> 

[buttercupck]

I ran into this exact issue yesterday. When I hit my weekly Claude limit and
tried to switch to GPT, I discovered that much of the infrastructure still relies
on the .claude directory structure. Making it truly LLM-agnostic would require
a significant refactor.

Regarding macOS voices: I also set up the native macOS voice integration
yesterday. Even with the Enhanced voices, the quality doesn't come close to
ElevenLabs. macOS definitely feels like a robot is greeting me vs 11labs hyper realistic voices.

[smolcompute]

Oh interesting. I thought all .claude/ had been replaced by $PAI_DIR. Is it just in the voice server ?

[Xantoshi-BTC]

No, there are a handfull stil in the code, in /hooks/ I've updated the files manually and made them dynamic. Happy to share if required.

[buttercupck]

Manually? Did your PA have trouble finding and changing those instances? I haven't had chance to tackle mine. Have you tested it out with other models?

[Xantoshi-BTC]

yes I am using Codex + MapleAI (local proxy)

~/.codex/config.toml

model_provider = "local_proxy"
model = "qwen3-coder-480b"
[model_providers.local_proxy]
name = "LocalProxy"
base_url = "http://127.0.0.1:8081/v1"
env_key = "OPENAI_API_KEY"

[parnoldx]

This looks promising. https://kyutai.org/next/tts

[smolcompute]

Found this too: https://voice-mode.readthedocs.io

[danielmiessler]

Hey all, I switched back to ElevenLabs voices because I realized they weren't actually charging me tons of money (it was my monthly charge for the whole year).

The MacOS voices just weren't good enough (expressive) enough for me, so I had to go back.

I hear you all on the privacy side.

Yes, we will be going heavy LOCAL and PRIVATE in the future.

Right now we're focused mostly on functionality and capability. But I HEAR YOU 100.