The candles are being lit…
All shall be well, and all shall be well, and all manner of thing shall be well.
Ignatius runs on modest hardware that rests when no one is praying, and takes a moment to warm back up (up to a minute). While you wait, breathe slowly, low in your belly.
Still loading? Close the app and open it again.
Pray with your own material
Upload a retreat handout, a few scripture passages or a reading. Ignatius at Home makes a guided audio retreat from it: each day's reading, a reflection for the heart, and a deep dive into the passage, woven into a prayer in the Ignatian tradition. What is this?
Sign in
Sign in with your email to keep your retreats. Make them on your laptop, pray on your phone.
Then enter the code from the email
Type the code from the email here, or just click the link in the email. (In an app on an iPhone's Home Screen, use the code: the link opens in the browser instead.)
Open the email and click the link; this page signs you in.
Trouble? Paste the link instead
Press and hold the link in the email, choose Copy Link (don't open it), and paste it here. A link works only once.
Just want to try it? A guest session uses free models and free voices, and stays in this browser.
Sign in on your phone
Scan this with your phone's camera to sign in there at once. Using the app from your iPhone's Home Screen instead? Open it and type this code:
The code and the QR code work once, for ten minutes.
·
Keep your retreats
This guest session lives only in this browser. Add your email to turn it into an account with the same retreats, then sign in with that email on your phone.
My retreats
New retreatExamples
Ready-made retreats to listen to and pray, each with a painting for every day. Each card says how it was made. Your progress in them is your own, and you can remove one from your home page after opening it.
New retreat
Making your retreat
It's fine to close this page; your retreat keeps being made and will be here when you come back.
Before you begin
Research done for this retreat
How each day was made: the passage and notes taken from your document, the searches run for the deep dive, every result that came back (with the service that found it), and which of them the deep dive cites.
Talk it over
A spoken conversation about your retreat with an AI prayer companion. It knows which days you've prayed, what they were about, what you've told it about yourself, and what you talked about before. It mostly asks questions and helps you notice where God has been at work; it gives little advice.
Taking turns: just talk, and when you pause the companion answers after a moment in a free voice. Slower than the live voices, but it costs only what the brain reads and writes, so a conversation that is mostly you talking costs very little, and with the free brain nothing at all.
Your choice is remembered on this device. What you want from the companion is set under About me.
The companion's instructions (advanced)
The full instructions the companion follows, before what it knows about you and this retreat. Change them if you like; they're saved to your account and used in every conversation. The shared background (under New retreat → Advanced) is always added after them. Every agent's prompt, and the companion's model, are also on the Agents page.
Still there? Live voices cost about 5 cents a minute even in silence. For quiet prayer, the free voice that takes turns costs nothing while you're silent.
This is an AI, not a priest, spiritual director or counselor. If you're in crisis or thinking about harming yourself, call or text 988 (in the US) or your local emergency number.
Past conversations
The companion remembers these: the most recent in full, older ones as a short summary. They're saved to your account.
Settings
About me
Anything you'd like the app to know about you: your situation, what you're carrying, how you like to pray. It shapes how your retreats are written and how the conversation companion talks with you. It's saved to your account as user info.md and is private to you.
Listening
How fast the recorded voices speak: the readings, reflections, deep dives, practices and the companion when it takes turns. The pitch stays the same, and silences keep their length. Live voices (GPT-Live, Grok) speak at their own pace.
Hands-free works like a voice conversation with Claude: after the companion speaks, it listens, and it waits a little longer after just a word or two, or after "and…" or "um…". You can pause listening, send at once, or interrupt the companion with the button. Tap to talk is better somewhere noisy.
Your calendar
Every day you pray, and every practice you finish (like the Examen), can appear on your own calendar at the time you prayed it, from when you began to when you finished: Prayed · Week 3 · Day 1 · Consideration of the way things are, 10:42–10:58 p.m., with links to this app and to that day.
Google Calendar
- Open Google Calendar: add from URL
Paste the address into "URL of calendar" and choose Add calendar.
Or try the one-tap link: Add to Google Calendar. Google checks the feed for new days every few hours.
iPhone, iPad or Mac
To choose how often it checks for new days, on your iPhone open Settings → Calendar → Accounts → Fetch New Data (on iOS 18: Settings → Apps → Calendar → Calendar Accounts). A web page can't open that screen for you.
The address is private to you: anyone with it could see which days you prayed. stops the address working.
Apple Health
Count your prayer as Mindful Minutes in Apple Health on your iPhone or iPad. A web app can't write to Health itself, so a small Apple Shortcut does it: after each prayer or practice, tap Add to Apple Health and the real minutes are logged.
Make the Shortcut once
- Open the Shortcuts app and tap +. Name the shortcut Log Prayer.
- Add the action Get Numbers from Input. It should read "Get numbers from Shortcut Input".
- Add Adjust Date: Subtract Numbers minutes from Current Date (that's when you began).
- Add Log Health Sample: Type Mindful Minutes, Value Numbers, Unit min, Date Adjusted Date.
- Tap Done. If it doesn't receive the number, open the shortcut's details (ⓘ), turn on Show in Share Sheet, and set it to receive Text.
The first time, the Shortcut asks permission to write to Health: allow it. You can delete the test minute in the Health app (Browse → Mental Wellbeing → Mindful Minutes).
Your highlights
Select words in a passage, a reflection, a deep dive or on the prayer screen, then tap Highlight. They're kept here, on every device you sign in on.
One each week
One of your highlights, the one sent longest ago, comes back to you once a week.
Agents
Every AI agent in the app, from the planner to the prayer companion: what it does, the prompt it follows, and the model it uses. Read them, make your own versions, or choose a different model.
Open the agentsServer start-up
How long people wait for the server when they open the app.
The server sleeps after 15 quiet minutes (Render's free tier), and the next person to open the app waits while it starts: that is a cold start. Every visit records its wait, and a daily check at 5:17 a.m. Eastern adds one even on days nobody visits.
Account
Debug mode
Waiting animation
What the companion shows while it writes a reply in the typed modes. All five are playing below; choose one.
Eval analysis: how the AI judges scored the models, and whether they agree.
Eval analysis
How the AI judges scored the models' reflections, deep dives and companion conversations, and, before trusting any ranking, whether the judges agree with each other enough to be measuring anything.
Agents
Each part of the app is written or spoken by an AI agent following a prompt: a set of instructions. Here are all of them. Read any prompt; make your own version to change how that agent behaves for you (it's saved to your account and used from the next time the agent runs); or go back to the default at any time. Where it makes sense, you can also choose a different model.
See a diagram of every agent and the tools it has
Always added (read only)
These go to the agents along with their prompts. The background and house style are shared by every agent; the format rules ask for exactly what the app reads back, so they can't be changed here.
About Ignatius at Home
Ignatius at Home turns prayer material you already have (a retreat handout, scripture passages, a reading, images) into a guided audio retreat you can pray at home, one day at a time.
The Spiritual Exercises
Ignatius of Loyola (1491–1556), the founder of the Jesuits, wrote the Spiritual Exercises as a guide for leading someone through a path of prayer; Pope Paul III approved them in 1548. They are arranged in four "weeks," stages rather than calendar weeks: God's love and mercy, the life of Christ, his passion, and his resurrection. Each time of prayer begins by asking for a particular grace, often enters a scene from the Gospels with the imagination, and ends in a colloquy, speaking to God "as one friend speaks to another." Along the way you learn to notice consolation and desolation, the movements of the heart toward and away from God, which is the root of Ignatian discernment.
- What are the Spiritual Exercises? · IgnatianSpirituality.com (Loyola Press)
- Ignatian contemplation: praying with the imagination · IgnatianSpirituality.com
- The rules for discernment · IgnatianSpirituality.com
- The Examen, the daily review of the day · IgnatianSpirituality.com
- The text of the Spiritual Exercises · Christian Classics Ethereal Library
- Spiritual Exercises of Ignatius of Loyola · Wikipedia
What a retreat is
A retreat is time set apart for prayer. The full Exercises can be made in about thirty days of silence at a retreat house, but Ignatius also allowed them to be made in daily life: at home over several months, with a set time of prayer each day and regular conversations with a spiritual director. Many parishes and Jesuit ministries run retreats like this through the year, week by week, with a handout of scripture and readings for each week. Ignatius at Home is built for that rhythm, and a retreat can continue from week to week as a series.
- An Ignatian Prayer Adventure, a free online retreat · IgnatianSpirituality.com
- Praying Every Day: an online retreat · Creighton University Online Ministries
- Retreat (spiritual) · Wikipedia
Lectio divina
Lectio divina, "divine reading," is the old monastic way of praying slowly with scripture. In the twelfth century the Carthusian monk Guigo II described its steps as a ladder: lectio, reading the text; meditatio, pondering what it says to me; oratio, answering God in prayer; and contemplatio, resting in God. Pope Benedict XVI's Verbum Domini (2010) adds actio, carrying the word into life.
- Verbum Domini, paragraphs 86–87 · Holy See
- Lectio divina · Saint John's Abbey, Collegeville
- Guigo II and The Ladder of Monks · Wikipedia
- Lectio Divina · Wikipedia
How a day is prayed here
- Ask for the grace of the day, then a short silence.
- First reading: simply listen (lectio).
- For the heart: a reflection that stays with the passage and invites you to notice what stirs.
- Second reading: listen for what it says to you (meditatio).
- Deep dive: the passage's setting, words, and how the church has read it.
- Third reading: listen for what God may offer or ask.
- Silence between two bells, resting with a word (contemplatio).
- Last reading, then speak to God in your own words (oratio, the colloquy), and close.
The voices
Every part of a day is read aloud by a synthetic voice, and you can choose a different voice for the guide, the reading, the reflection and the deep dive. Each sample below reads the same lines: John 20:16 from the World English Bible, then a line of spoken guidance.
Jesus said to her, "Mary." She turned and said to him, "Rabboni!" which is to say, "Teacher!" Take a moment now. Let the name spoken to you settle in your heart, and notice what stirs.
Free: Microsoft neural voices
Read through Microsoft's Edge "Read aloud" service (the open-source edge-tts package). They cost nothing and need no key, are clear and even, and handle scripture well. Their weaknesses: the pacing is steadier and less expressive than a human reader, long scripts are recorded in 400-character pieces so the tone can shift a little between sentences, and because the service is unofficial it occasionally drops a request (the app retries, with backoff).
Premium: ElevenLabs
ElevenLabs voices breathe, slow down and lower their voice the way a person does when reading something that matters, and they hold one tone across a whole reflection (the app sends 2,500-character pieces with the text before each piece, so the joins are seamless). For prayer that difference is real: a reflection meant to be listened to in silence sounds less like an announcement and more like someone sitting with you. They cost money: about 10,000–15,000 characters for a day, charged against the account's monthly credits, so they are for premium accounts; everyone can hear them in the premium example retreat.
Conversation voices
Talk it over uses live speech-to-speech models rather than recorded audio. OpenAI GPT-Live connects your browser straight to OpenAI over WebRTC, with very low delay and natural turn-taking; it costs about 5¢ a minute. xAI Grok voice streams audio over a WebSocket with a short-lived key from the server, has a choice of distinct voices (Eve, Ara, Rex, Sal, Leo), and costs about 8¢ a minute. The server never sends your key-holding credentials to the browser, only a single-use session.
Hear the conversation voices
Each says the same line: "Hello. I'm glad you're here. What stayed with you from today's prayer?" These were recorded with each company's text-to-speech. In a live conversation the same voice sounds a little more relaxed and conversational. OpenAI's other live voices (Vesper, Willow, Quartz, Meridian, Stone, Gleam, Beacon, Delta, Cinder and Ripple) are only available live, so they can't be sampled here; start a conversation to hear them.
How the companion talks: live voices and taking turns
Talk it over can work in two quite different ways. The live voices (OpenAI GPT-Live and xAI Grok) are one model that hears and speaks. Taking turns chains three separate steps: your device turns speech into text, a text model (the "brain") writes a reply, and a synthetic voice reads it aloud. Each is better at something.
Live voice speech to speech
Taking turns speech → text → model → text → speech
Why the live voices feel so human
A live voice model works on sound itself. Your voice is streamed to it continuously (over WebRTC for OpenAI, a WebSocket for Grok), cut into small pieces of audio the model reads much as a text model reads words, and it answers with pieces of audio, streamed straight back. Nothing in between turns your speech into words and back. So it hears more than your words (hesitation, a catch in the voice, a laugh), and it can answer with more than words: the small "mm" and "mhm" of someone listening, a breath before a hard sentence, a softer tone when you sound tired. None of that was written down; it comes from the model having learned from real speech.
It can be interrupted because it's always listening. The service watches the incoming sound for speech (voice activity detection). When you start talking while it's speaking, it stops, throws away the rest of what it was going to say, and listens; when you pause for a moment, it takes that as its turn. That is what makes it the most conversational way to talk with a model.
What the live voices give up
- Depth of thought. A speech-to-speech model has to start answering almost at once, so it doesn't stop to think the way a reasoning text model can, and it is generally not as strong a writer or reasoner as the best text models. Long instructions, your notes and the whole retreat are harder for it to keep in view.
- Tools. The text models behind the rest of the app can be given tools, such as the web search the deep dive uses to check facts. The live voices here have none: they can only talk. (Their services can hand hard questions to a separate text model with tools, but that adds delay, and this app doesn't use it.)
- Exact words. The transcript is the service's own guess at what was said, and following written instructions to the letter is harder for it.
- Cost while quiet. A live conversation is billed for every second it's open, silence included: GPT-Live costs about five cents a minute whether or not anyone speaks.
What taking turns does better, and what it gives up
Taking turns is how most AI assistants you type to become voice assistants: speech to text, the model, then text to speech. Because the middle step is an ordinary text model, you can choose the strongest writer (Claude Fable or Opus) or a free one; it reads the companion's full instructions, your notes and the retreat day by day, and can think before it answers; and it costs only for what it reads and writes, so silence is free. The price is the conversation's texture. It waits until you've finished; you can't cut in mid-sentence; your tone of voice is lost when your speech becomes text; and the reply is read by a synthetic voice, which pronounces words well but won't say "mm" unless the words say so.
A practical rule: for a flowing conversation, choose a live voice; for silence-heavy prayer, or for the most thoughtful replies, take turns. Choose under Change voice on Talk it over, or on the Agents page.
The research services
With the free open models, the deep dive is written from web research the server does first: the model proposes three searches, the services run them, and the model writes from what comes back, citing only pages that were actually returned. Claude models get the same results as a head start and then search the web themselves as much as they need. By default all services, combined are asked at once and their results are interleaved, one from each in turn, so no single index decides what the model reads. If a service runs out of monthly credits it's paused until next month and the others carry on; every query and result is logged. On a computer, the research behind each day is under "Research notes" on the retreat page.
- Brave Search
- Brave's own independent web index (not a reseller of Google or Bing). Fast, broad, and good for mainstream reference pages. The snippets are short, so the model sees less of each page.
About 1,000 searches a month free ($5 of monthly credit). Get a key · Quickstart · Web search reference · Pricing · key:BRAVE_SEARCH_API_KEY - Brave Answers
- An AI-written answer from the Brave index, with its citations. A quick summary with sources in one call; but the answer is itself AI-generated, so it's one step removed from the pages.
$5 of monthly credit free; a separate plan to switch on, and its own key, in the same Brave account. Get a key · Answers guide · Answers reference · Pricing · key:BRAVE_ANSWERS_API_KEY - Exa
- A search engine built for AI that searches by meaning rather than keywords, and returns highlighted passages. Strong at finding essays, commentaries and scholarly writing on an idea; weaker for exact phrases and very recent news.
About 1,400 searches a month free ($10 of credit, reset on the 1st; no card). Get a key · Quickstart · Search reference · Pricing · key:EXA_API_KEY - Tavily
- A search API made for AI agents that returns cleaned, relevance-ranked extracts of each page. Good all-round results with useful content; its free tier is a modest monthly number of searches.
1,000 searches a month free (no card). Get a key · Quickstart · Search reference · Pricing · key:TAVILY_API_KEY - Firecrawl
- Mainly a tool for reading web pages; its search returns results with page content. Useful for getting the substance of a page; slower than the others, and free credits go quickly.
About 500 searches a month free (1,000 credits; no card). Get a key · Introduction · Search reference · Pricing · key:FIRECRAWL_API_KEY - Linkup
- A search API with a standard mode and a deep mode that researches a question over several steps and returns a sourced answer. Deep research is thorough but takes about 20 seconds a question, so it's offered only on its own, not in the combined mix.
$20 of credit a month free: about 4,000 standard or 360 deep searches. Get a key · Quickstart · Search reference · Pricing · key:LINKUP_API_KEY
To run your own copy, you don't need all six: any one is enough, and with none the deep dive is written without web research. Sign up (each has a free tier), create an API key, and put it in the server's environment under the name shown: in .env when running it on your computer, or under Environment on Render. The server uses every service that has a key. Setup is in the backend README. Free allowances are as the services listed them in September 2026.
The spoken guidance, reflection and deep dive are written by AI models from your material, and the voices are synthetic. Treat them as a companion to your own prayer and, if you have one, your spiritual director, not a replacement. Upload only material you have the right to use; your retreats are private to your account.
A FastAPI backend on Render; this page on GitHub Pages. How it works.
Costs
What each retreat cost to make, by part and by company. Nothing here is shown anywhere else in the app.