The AI companion
Umbra, the AI companion in Soul Arc
Umbra is the AI companion inside Soul Arc, a shadow work journal for iPhone. Umbra reads what you write and replies with short, direct questions. It does not give advice, and it does not diagnose.
Updated October 2026What is Umbra?
In the app, Umbra is called The Witness. It appears as a small skeleton character, and it talks with you in two places:
- Speak with Umbra. A short conversation you start from the Shadow tab.
- Deep Dive. A follow-up conversation about one thing you wrote, such as a journal entry or a Shadow Flare.
Umbra is not a person and not a therapist. Its replies are written by Claude Sonnet 4.6, an AI model made by Anthropic.
Umbra is part of Soul Arc. It is not connected to other apps or products that use the name Umbra.
How does a session with Umbra work?
| Free | Pro | |
|---|---|---|
| Speak with Umbra sessions a day | 1 | 10 |
| Messages you can send per session | 5 | 10 |
| Deep Dives a day | 1 | 10 |
| Turns per Deep Dive | 5 | 10 |
A session has a shape:
- Umbra opens with one line. The app picks it based on whether this is your first session, how long you have been away, and your streak.
- Early on, Umbra stays brief. One observation or one question.
- In the middle, it goes under the surface. It names patterns, contradictions, and things you seem to be avoiding.
- Near the end, it asks one direct question.
- Its last message is a single statement about what it saw. It does not wrap things up for you.
Every reply is short. The app asks the model for at most 200 tokens, which is a few sentences.
What will Umbra not do?
Umbra's instructions set these limits:
- It never gives advice, makes suggestions, or tells you what to do.
- It never diagnoses you or uses clinical language.
- It does not offer silver linings, and it does not praise you for opening up.
- It does not analyze other people. It is told to look only at you.
- It does not discuss politics, religious doctrine, or current events.
These are instructions, not guarantees. An AI model can still get things wrong, so treat what Umbra says as a question to think about, not as a fact about you.
Does Umbra remember me?
A little. When a session ends, the app makes a short title and summary of it. At the start of your next session, Umbra is given the summaries of your last 5 sessions. This works the same on the free plan and on Pro.
What leaves my phone when I talk to Umbra?
Your journal as a whole is never uploaded. It is stored on your device, and there is no account. Each request to Umbra carries only this:
| What is sent | How much |
|---|---|
| Your messages in this session | Up to 500 characters each |
| The name you gave Umbra | Up to 30 characters |
| Your progress | Archetype, awareness stage, current planet, streak, days since your last session, planets unlocked |
| Your 3 most recent entries | The date, the first 100 characters of the prompt, and the first 150 characters of what you wrote |
| Your 3 most frequent Shadow Flares | The first 50 characters of each, and your average intensity |
| Past sessions | Summaries of your last 5 sessions |
| In a Deep Dive | The full text of the entry or session you chose |
| Identifiers | A random device ID and an anonymous purchase ID, used to count daily limits |
The request goes from the app to our server, which passes it to Anthropic's API to write the reply. Your email is never sent, because the app never asks for it.
What is kept on a server, and for how long?
| What | Kept on our server? |
|---|---|
| Your messages and journal excerpts | No. They pass through to Anthropic and are not stored by us. |
| Umbra's reply | For up to 30 minutes, so a dropped connection can retry without using another turn. Then it is deleted. |
| A device ID and a time for each request | For up to 30 minutes, to stop bursts of requests. |
| Daily usage counts, whether Pro is active, and an anonymous purchase ID | Yes, tied to the random device ID. Email us to have them deleted. |
| If the AI service returns an error | The error type and the device ID. The device ID is removed after 90 days. |
Anthropic processes what it receives in order to write the reply. Anthropic says it does not use content sent through its API to train its models. The privacy policy has the full picture, including how to ask for deletion.
What happens if I write about suicide or self-harm?
The app checks each message on your phone for phrases about suicide or self-harm. If it finds one, it pins a card to the top of the conversation with two lines to reach: the Suicide and Crisis Lifeline (call or text 988) and the Crisis Text Line (text HOME to 741741).
The conversation stays open, and a session that ends soon after this does not count against your daily limit. Umbra's instructions also tell it to stop the usual work and point you to those same lines.
Umbra is not a crisis service. If you are in danger, call or text 988 in the US now.
Questions people ask
Is Umbra a real AI, or is it scripted?
Both, in different places. Replies in Speak with Umbra and Deep Dive are written by an AI model. Umbra's opening line in Speak with Umbra, the journal prompts, and the short lines Umbra says elsewhere in the app are written by hand.
Is Umbra a therapist?
No. Umbra is an AI character in a journaling app. It cannot assess, diagnose, or treat anything, and it is not a substitute for professional mental health care.
Why are sessions with Umbra so short?
They are short on purpose. A free session is 5 messages and a Pro session is 10. Umbra's last message is a closing statement, not another question.
Can I read my past conversations with Umbra?
Yes. Finished sessions are saved on your device and listed under Previous Sessions on the Shadow tab.
Which AI model does Umbra use?
Speak with Umbra and Deep Dive use Claude Sonnet 4.6. Session titles and summaries use Claude Haiku 4.5. Both are made by Anthropic.