GUIDING LIGHT

A space to think out loud.

Sign in to continue.

LIVE CONVERSATION

Subscribe for live turns.

Scripted demo is free with sign-in. Live conversation needs a plan: each includes hours of live voice per month, and the bigger plans cost less per hour.

gguiding lightVOICE CONVERSATION LAB

YOUR SPACE

A LITTLE ROOM TO PAUSE

You don’t have to arrive
with the right words.

◈ Keys stay on your server

One continuous conversation: Guiding Light remembers what you said, and when, across visits and devices. Manage it under Memory. Live requests are sent to your chosen AI providers.

Your space / Conversation

Connecting

ONE THOUGHT AT A TIME

A space to think out loud.

Start wherever you are. There’s no perfect way to begin.

GUIDING LIGHTGetting ready

What’s on your mind?

Your microphone stays on. Speak to begin; pause to send.

AI-generated voice · Not a human therapist

DEMO MODE A scripted reply lets you test the interface without API keys. It is not a live AI conversation.

Conversation

A fresh start

Your conversation will appear here.
Use your voice, or start with a few words below.

Your message

Enter to send · Shift + Enter for a new line

Speech → Context + history → Meta + reply → Voice

⌘ Session inspectorMetadata is never spoken

Generated session notes are tentative summaries, not diagnoses or the model’s hidden chain-of-thought. Visible here for development.

Transcription—

First text token—

Speech field ready—

First audio scheduled—

No metadata yet.

An experimental AI reflection tool, not clinical care or an emergency service.
In immediate danger, contact local emergency services or someone you trust.

BETWEEN CONVERSATIONS

Memory

What Guiding Light carries into your next conversation, updated after every message. Working memory is the short summary it always has in mind; edit it freely and your version wins. Below it are your sessions: every message is kept word for word and summarized over time, so it can look back for detail. Delete anything, anytime.

Working memory

Sessions

MAKE IT YOURS

Prompt & settings

API keys belong in the backend.

Add keys to .env and restart npm start. This page never receives them.

Scripted demo — no paid model calls

Voice pipelineCosts are estimates for a typical hour.Live voiceGPT-Live speaks in this voice.

Voice directionHow the voice should sound: pace, warmth, tone. Used by GPT-Live and the OpenAI voice.

Method profileResponse timingResponsive starts speaking while reflection and memory run in the background.

Reflection word budgetIn Responsive mode this limits background notes. Memory remains active at zero.

Text modelReasoning effortOnly efforts the selected model supports are offered.Transcription modelSpoken language

Speech modelVoice

ElevenLabs modelElevenLabs voiceVoices from your ElevenLabs account.

Cartesia modelCartesia voiceVoices from your Cartesia account.

Include the previous meta as tentative continuity notes Your conversational promptThe backend adds the XML envelope and basic AI/safety boundaries separately. Context about the current userSent alongside backend-owned dialogue history on every live turn.

Your supplied <meta> style example

Use the example as a style reference, not active user facts

This contains the personal example supplied with the build. Remove or anonymize it before sharing the project.

Settings are held by the server.