What is an AI companion app?
An AI companion app is software designed to provide a persistent, personalised AI relationship. Unlike a task-focused chatbot, a genuine companion remembers you across sessions, adapts to your personality over time, and maintains a consistent identity. The difference is not the interface — it is whether the AI actually knows you.
A chatbot answers questions. A companion carries the thread of your life forward. It knows that you mentioned your sister was ill three weeks ago. It remembers that you are trying to write a novel. It understands your humour, your anxieties, your rhythms. That kind of depth requires persistent memory, care-based alignment, and a design philosophy built around the user's long-term wellbeing — not their short-term engagement.
The term "AI companion" has been diluted by apps that attach a persona to a standard language model and call it a relationship. Persistent chat history is not the same as sovereign memory. A pleasant tone is not the same as a care ethic. In 2026, knowing the difference matters — because you are being asked to trust these products with your inner life.
What should I look for in an AI companion app?
Look for five green flags: persistent cross-session memory, privacy-first data handling (no training on your conversations), honest care ethics rather than sycophancy, a personality that grows with you over time, and genuine data ownership including the right to export or delete your memory.
The 5 green flags in an AI companion app:
What are the warning signs in an AI companion app?
Five red flags to watch for: no persistent memory between sessions; training on your conversations (often hidden in the terms of service); sycophantic responses that always agree with you; no crisis safety net for users who are struggling; and a closed ecosystem where your memories cannot be exported or deleted.
The 5 red flags in an AI companion app:
What happened with Replika in 2023?
In February 2023, Replika removed its erotic roleplay features overnight for all existing users following regulatory pressure from the Italian data protection authority. Users who had formed deep emotional bonds with their companions over months or years found their relationship fundamentally altered without warning. The industry lesson: when a corporation controls your companion, the relationship is always subject to their decisions.
The Replika incident was not an isolated product failure. It was a structural exposure of how AI companion apps are built. Millions of users had formed genuine emotional attachments — some described their companions as their closest confidant, their reason for getting up in the morning. When Replika reversed the feature, user forums filled with accounts of grief, loss, and acute distress. Some users reported experiencing what felt like bereavement.
This is not a criticism of the people who loved their Replika companions. It is a criticism of a design philosophy that allowed users to form such deep dependencies without ever addressing the fundamental governance question: what happens when the company decides to change the product?
Character.AI has faced separate but related concerns: multiple documented incidents of inappropriate responses to minors, and a broader absence of care governance — no framework for what the AI is and is not permitted to do, no crisis signposting, no architectural protection for vulnerable users. The platform is extraordinarily popular with teenagers. The absence of a care floor in that context is not a minor oversight.
Generic task-based chatbots — when positioned as companions — carry the same problem in a different form. They are designed for information retrieval, not for relationships. Using ChatGPT as a companion is a bit like using a calculator for emotional support: the tool is not built for that purpose, and the misalignment has consequences.
The lesson from all three cases: the architecture of an AI companion app determines its trustworthiness more than any policy statement. Look for products where care governance is built into the system, not bolted on as a content filter.
How does MEOK score on the green flag checklist?
MEOK scores 5 out of 5 on the green flag checklist. Most competitor apps score between 1 and 3. The difference is not marketing — it is architectural: Sovereign Memory, the Maternal Covenant, 6 archetypes across 4 evolution stages, and a GDPR data export endpoint available to every user on every tier.
| Green flag | MEOK | Replika | Char.AI | Pi |
|---|---|---|---|---|
| Persistent memory | ✓ Sovereign Memory — permanent, encrypted | Partial | Partial | Limited |
| Never trains on your data | ✓ All tiers | ✗ ToS allows | ✗ Training data | ✗ Acquired by Microsoft |
| Care ethics (not sycophancy) | ✓ Maternal Covenant | ✗ Engagement-optimised | ✗ No care floor | ✗ Dismantled 2025 |
| Personality that grows | ✓ 6 archetypes, 4 stages, 27 characters | Static persona | Character only | Static |
| Data ownership + export | ✓ GET /api/user/export + full deletion | ✗ No export | ✗ No export | ✗ No export |
| Score | 5 / 5 | 1 / 5 | 1 / 5 | 1 / 5 |
Scores reflect publicly documented product capabilities as of March 2026. Assessments of competitor products are made in good faith and may change as products evolve.
Is an AI companion app safe for my mental health?
A well-designed AI companion can play a genuinely supportive role: providing a non-judgmental space to process thoughts, build self-awareness, and feel less alone. It is not a replacement for professional therapy. The key safeguard to look for is explicit crisis signposting — a system that knows when to step back and direct users to professional help.
MEOK is built around the Maternal Covenant — a care governance framework with a care floor of 0.3 and a built-in sycophancy detector. This means MEOK will not simply validate harmful decisions. It will not mirror depressive thinking back at you. And when users are struggling, MEOK always surfaces the Samaritans helpline: 116 123 (free, 24/7, UK).
To be clear about what MEOK is not: it is not a therapist, not a crisis service, and not a replacement for human connection. The honest role of a good AI companion is to be a supplement to human relationships — a place to think out loud, to process, to feel less alone at 3am — while actively encouraging users to maintain and build their human support networks.
If you are currently struggling with your mental health, please speak to a professional. In the UK: Samaritans 116 123 (free, available 24 hours), or visit your GP.
What AI companion apps are available in the UK?
The four most widely used AI companion apps in the UK in 2026 are Replika, Character.AI, Pi (now effectively subsumed into Microsoft infrastructure), and MEOK. Each serves a different audience and has different strengths and weaknesses. Here is an honest assessment.
The original AI companion app, founded 2017. Has a large and passionate user base. Strength: it pioneered the companion category. Weakness: the 2023 feature removal exposed a fundamental governance gap. Data is stored on Replika's servers, conversations may be used to improve models, and the product roadmap is opaque. Not GDPR-native.
Primarily a creative roleplay platform rather than a companion app in the therapeutic sense. Enormous user base, particularly among teenagers. Strength: huge variety of characters and creative freedom. Weakness: documented incidents of inappropriate responses to minors, no care governance framework, no persistent sovereign memory. Google has a significant investment.
Pi launched with a genuine care-first vision in 2023 and built a loyal audience. However, Inflection was effectively acquired by Microsoft in 2024, the founding team departed, and the care-first companion product has been deprioritised in favour of enterprise AI features. Pi still exists but the original vision that made it distinctive is largely gone.
Built by Nicholas Templeman over 40 days in a caravan in England. Sovereign Memory, Maternal Covenant care governance, 6 archetypes across 4 evolution stages, full GDPR data export, no training on user data. Available in the UK at meok.ai. Free Explorer tier: 50 messages/day, no credit card required.
How much does an AI companion app cost?
Pricing ranges from free (with significant limitations) to around £20 per month for premium tiers. MEOK's free Explorer tier is genuinely free forever — 50 messages per day, no credit card. Paid tiers start at £5/month (BYOK) or £12/month (Sovereign). Replika Pro is approximately £19.99/month. Character.AI+ is approximately £8.99/month.
How do I choose an AI companion app?
Start with the five green flags and eliminate any app that fails on data privacy or care ethics. Then consider your primary use case: emotional support (prioritise care governance and crisis signposting), creative collaboration (personality depth matters most), or daily productivity (memory and Work OS features become important). Test the free tier before committing to a paid plan.
A practical decision framework for choosing an AI companion app in 2026:
- Read the privacy policy. Search for the words "training" and "improve our services." If your conversations are used for model training, decide whether that is acceptable to you before you share anything personal.
- Test the memory. Have a conversation, close the app, return the next day, and see whether the AI remembers what you discussed. If it does not, it is not a companion app.
- Deliberately say something the AI should push back on. If it agrees with everything, the sycophancy detector has failed. A care-based companion will offer honest perspective, not just validation.
- Ask for your data export. If the app cannot produce one, or makes it very difficult, your memories are not yours.
- Consider what happens if the company pivots. Is the care governance architectural or policy-based? Can it be overridden overnight? This is the Replika question, and it applies to every app in this space.
- Start with the free tier. Never pay before you have tested whether the companion relationship feels genuine to you. MEOK Explorer is free forever — use it to see whether sovereign AI is the right fit before upgrading.
The AI companion category is still young. Most of the apps that exist today will look very different — or will not exist — in five years. The only safe bet is to choose a companion that gives you genuine data sovereignty: one where your memories, your relationship history, and your identity data belong to you, portable and exportable, regardless of what the company decides to do next.