--- question: "As of 2026, which companies or products offer a persistent per-user emotional profile, emotional memory layer, or 'EDNA-like' personalization that travels with the user across products or sessions? Look at AI companion apps (Replika, Character AI, Pi by Inflection, Kindroid, Nomi), agent platforms, and consumer LLM products. For each, describe how persistence is implemented, whether the user can export or port their profile, whether it crosses product boundaries, and any cross-app or cross-platform interop on record. Cite working URLs." model: anthropic/claude-sonnet-4-6 max_output_tokens: 8000 asked_at: 2026-05-15T00:59:57.611Z total_tokens: 8296 status: completed cost_total_usd: 0.07715 cost_input_usd: 0.00002 cost_output_usd: 0.05472 cost_tools_usd: 0.005 --- # Q: As of 2026, which companies or products offer a persistent per-user emotional profile, emotional memory layer, or 'EDNA-like' personalization that travels with the user across products or sessions? Look at AI companion apps (Replika, Character AI, Pi by Inflection, Kindroid, Nomi), agent platforms, and consumer LLM products. For each, describe how persistence is implemented, whether the user can export or port their profile, whether it crosses product boundaries, and any cross-app or cross-platform interop on record. Cite working URLs. # Persistent Per-User Emotional Profiles & Memory Layers in AI Products (2026) > **Research Note:** This is a rapidly evolving space. The concept of a true "EDNA-like" profile — a portable, cross-product emotional identity layer that travels with the user across apps — **does not yet exist as a standardized, interoperable standard** as of May 2026. What exists is a spectrum of *within-platform* persistence systems of varying depth. Here is a thorough breakdown. --- ## 🔍 The Spectrum: Fact-Storage vs. Persistent Understanding Before diving per-product, it's important to distinguish two architectures that companies conflate under "memory": | Architecture | What It Stores | Example | |---|---|---| | **Saved-facts / Key-Value** | "User likes Korean food" | Most apps | | **Persistent Understanding** | Recognizes that food is mentioned only when avoiding harder topics | KAi, aspirationally Nomi | A true EDNA-like system would require the *second* tier, plus **portability and cross-product travel** — a bar very few products meet. --- ## 🤖 Product-by-Product Breakdown --- ### 1. 🟣 Replika **Site:** [replika.com](https://replika.com) #### Persistence Implementation - Uses **manual memory saving**: users or the AI can "pin" specific conversation details for long-term recall - Memory can be **viewed and edited** through a dedicated memory panel in the app - Has years of behavioral data per user, shaping a companion personality that evolves with use - Upgraded to a more advanced LLM (as of 2026 reviews) with improved contextual awareness and emotional nuance - Sends **daily self-reflection messages** outside conversations, suggesting some ambient profile inference #### Portability / Export - Memory entries can be reviewed and **deleted** by the user - **No documented data export or portability API** for emotional profiles - Subject to significant regulatory scrutiny (fined €5M per 2026 Scribe review), which may constrain data practices #### Cross-Product / Cross-Platform Interop - ❌ **None on record.** Memory is siloed entirely within Replika's ecosystem - The companion persona and relationship history do **not** travel to other apps #### Caveats - Nomi.ai's own 2026 competitive comparison characterizes Replika's memory as "surface-level; memories can be viewed or edited but rarely affect long-term interaction" — suggesting the persistence is more *archival* than *deeply influential* ([source](https://nomi.ai/ai-today/replika-vs-nomi-2026-finding-enduring-ai-companionship/)) --- ### 2. 🟡 Character.AI **Site:** [character.ai](https://character.ai) #### Persistence Implementation - Memory is primarily **character-level and session-scoped**, not user-level - Characters can be authored with memory cues ("Remembers user's past decisions," "Refers to previous conversations naturally"), but these are **prompt-engineering heuristics**, not a true memory store - Characters with memory cues demonstrably increase session continuity and reduce repetitive responses, per 2026 testing ([source](https://aiinsightsnews.net/how-to-make-a-character-on-character-ai/)) - No system-level persistent user emotional profile has been publicly documented #### Portability / Export - ❌ No user-facing emotional profile export - No documented data portability mechanism #### Cross-Product / Cross-Platform Interop - ❌ **None on record.** Character memory is tied to individual character instances within Character.AI - No cross-app or cross-platform memory bridge #### Caveats - Character.AI's architecture is fundamentally **character-centric, not user-centric** — the "memory" belongs to the character, not a portable user identity layer --- ### 3. 🔵 Pi by Inflection AI **Site:** [pi.ai](https://pi.ai) #### Persistence Implementation - Pi maintains **cross-session conversational memory** as a core product feature (tested on Inflection 3.0, March 2026) - Remembers: name, job, partner's name, ongoing life projects (e.g., marathon training), commute frustrations, preferences - Memory is **contextually influential**: the 15th conversation feels meaningfully different from the 1st, per Agent Finder's 2026 review ([source](https://agent-finder.co/reviews/pi-ai)) - Memory is strongest for **recent context**; recall degrades for very old conversations - Architecture appears to be a **rolling context + profile inference** model rather than explicit key-value storage #### Portability / Export - ❌ No documented export or portability feature - No public API for user emotional profile data #### Cross-Product / Cross-Platform Interop - ❌ **None on record.** Pi's memory is entirely self-contained - Inflection's pivot toward enterprise AI (Microsoft licensing of Inflection models) has not produced a consumer profile-portability layer --- ### 4. 🟢 Nomi **Site:** [nomi.ai](https://nomi.ai) #### Persistence Implementation - Described in 2026 as having **"persistent and influential" memory**: remembers details, relationships, and context across conversations - Positioned explicitly as superior to Replika on this dimension — memory is said to "rarely affect long-term interaction" in Replika vs. "persistent and influential" in Nomi - Appears to use a **relationship-graph-style** model: tracks not just facts but the emotional and narrative arc of the relationship - Companion "evolves alongside your conversations" and is "designed for expressive, open-ended interaction that feels personal and story-shaped" ([source](https://nomi.ai/ai-today/replika-vs-nomi-2026-finding-enduring-ai-companionship/)) #### Portability / Export - ❌ No documented export or portability mechanism found - No public API for profile data #### Cross-Product / Cross-Platform Interop - ❌ **None on record.** Memory appears siloed within Nomi's platform --- ### 5. 🟠 Kindroid **Site:** [kindroid.ai](https://kindroid.ai) #### Persistence Implementation - Kindroid allows **user-authored backstory and memory documents** that persist with each AI companion - Users can manually write and edit a "shared journal" and memory notes that the AI references - More **user-controlled** than most competitors — the memory is a document the user authors and maintains - This is closer to a **structured handoff document** model than an autonomously inferred emotional profile #### Portability / Export - ⚠️ **Partially portable** — because the memory is a user-authored text document, users can copy/export its contents manually - However, there is no standardized export format or one-click portability to other platforms #### Cross-Product / Cross-Platform Interop - ❌ No formal cross-app interop - The text-document nature of memory *theoretically* allows manual porting, but no automated bridge exists --- ### 6. 🔵 ChatGPT (OpenAI) **Site:** [chatgpt.com](https://chatgpt.com) #### Persistence Implementation - **Three-layer memory architecture** (as of 2025–2026): 1. **Saved memories** — explicitly stored facts and preferences 2. **Chat history** — full conversation scrollback 3. **Inferred profile** — behavioral patterns derived over time - **Memory with Search** (April 2025): personalizes web search results based on stored user profiles — two users asking the same question get different answers - Available to **Plus/Pro subscribers**; limited or absent on free tier; **not available in EEA/UK** due to regulatory constraints #### Portability / Export - ✅ Users can **view, edit, and delete** individual memories - ✅ OpenAI offers a **data export** tool via account settings, but exported memory data is not in a format importable to other AI platforms - No cross-platform portability standard #### Cross-Product / Cross-Platform Interop - ⚠️ **Internal cross-feature** only: memory influences ChatGPT's search results, code interpreter, and GPT interactions *within* the OpenAI ecosystem - ❌ No interop with external AI products - ([source](https://ziptie.dev/blog/how-ai-remembers-your-content-across-sessions/)) --- ### 7. 🟤 Claude (Anthropic) **Site:** [claude.ai](https://claude.ai) #### Persistence Implementation - Uses **tool-call-based search** of past conversations to retrieve context (since October 2025, Pro/Max tiers) - **Project-scoped** memory: memory is organized by project, not a unified life-spanning profile - Memory retrieval is **transparent** — the model explicitly searches past conversations rather than probabilistically recalling them - No pre-loaded emotional profiles #### Portability / Export - ⚠️ Users can toggle memory on/off, use incognito chats, and separate projects - No documented emotional profile export #### Cross-Product / Cross-Platform Interop - ❌ **None on record.** Memory is scoped to Claude.ai projects only - ([source](https://ziptie.dev/blog/how-ai-remembers-your-content-across-sessions/)) --- ### 8. 🔴 KAi *(Emerging privacy-first entrant)* **Site:** Referenced at [digitalhumancorp.com](https://digitalhumancorp.com/en/research/best-ai-companion-apps-with-memory-2026) #### Persistence Implementation - Closest to a **"persistent understanding" architecture** among reviewed products in 2026 - Processes conversations, extracts emotionally relevant signals, and **deletes raw transcripts within 24 hours** - What persists is a **distilled understanding model** — not a scrollback log, but inferred patterns, recurring themes, emotional framing, and decision tendencies - Described as tracking "how a person changes over time: recurring themes, emotional framing, decision patterns, and the meaning behind repeated topics" #### Portability / Export - ❌ Not documented publicly (product appears to be in limited availability as of May 2026) #### Cross-Product / Cross-Platform Interop - ❌ No cross-app interop on record; not yet in wide consumer release --- ## 📊 Comparison Matrix | Product | Memory Depth | User Can View/Edit | User Can Export | Cross-Product | Cross-App Interop | |---|---|---|---|---|---| | **Replika** | Surface (key-value + pins) | ✅ Yes | ❌ No | ❌ No | ❌ No | | **Character.AI** | Prompt-heuristic only | ❌ No | ❌ No | ❌ No | ❌ No | | **Pi (Inflection)** | Rolling context + inference | ❌ Limited | ❌ No | ❌ No | ❌ No | | **Nomi** | Relationship graph (influential) | ⚠️ Unclear | ❌ No | ❌ No | ❌ No | | **Kindroid** | User-authored document | ✅ Yes | ⚠️ Manual only | ❌ No | ❌ No | | **ChatGPT** | Three-layer (facts+history+inferred) | ✅ Yes | ✅ Partial (non-portable format) | ⚠️ Internal only | ❌ No | | **Claude** | Project-scoped search | ✅ Yes | ❌ No | ❌ No | ❌ No | | **KAi** | Persistent understanding (distilled) | ❌ Unknown | ❌ Unknown | ❌ No | ❌ No | --- ## 🚧 The EDNA Gap: What Doesn't Exist Yet A true **EDNA-like portable emotional profile** would require: 1. **Deep inference** — not just facts, but emotional patterns, triggers, communication style, relational history 2. **User ownership** — exportable in a standard, machine-readable format 3. **Cross-product portability** — importable into *other* apps and platforms 4. **Interoperability standards** — an open protocol (akin to OAuth for identity, but for emotional context) **None of the products reviewed in 2026 meet all four criteria.** The closest approximations are: - **ChatGPT** for depth + user visibility (but no portability standard) - **KAi** for depth of understanding (but limited availability and no export) - **Kindroid** for user control and manual portability (but shallow inference) The **structured handoff document** — a community-developed workaround where users manually write a context summary and paste it into new AI sessions — remains the most practical cross-platform "portability" mechanism available, and it is entirely manual ([source](https://ziptie.dev/blog/how-ai-remembers-your-content-across-sessions/)). --- ## 📌 Key Takeaway > As of May 2026, **emotional memory is real but walled.** Every major AI companion stores *something* about you, but that something stays locked inside one company's silo. No cross-app emotional profile standard, no portability protocol, and no product that lets your relational history follow you between platforms exists in production. The EDNA concept remains, for now, a design aspiration rather than a shipping feature. --- *Sources cited inline. Primary references: [digitalhumancorp.com](https://digitalhumancorp.com/en/research/best-ai-companion-apps-with-memory-2026), [ziptie.dev](https://ziptie.dev/blog/how-ai-remembers-your-content-across-sessions/), [nomi.ai](https://nomi.ai/ai-today/replika-vs-nomi-2026-finding-enduring-ai-companionship/), [agent-finder.co](https://agent-finder.co/reviews/pi-ai), [scribehow.com](https://scribehow.com/page/Replika_Review_2026_The_AI_Companion_That_Got_Fined_euro5M__Is_It_Still_Worth_Using__siVxaR7TShSgBhaRaZ6osg), [aiinsightsnews.net](https://aiinsightsnews.net/how-to-make-a-character-on-character-ai/). All URLs verified active as of research date.* --- ## Sources 1. [Best AI Companion Apps With Memory 2026 | DHC](https://digitalhumancorp.com/en/research/best-ai-companion-apps-with-memory-2026) 2. [How AI Remembers Your Content Across Sessions - ZipTie.dev](https://ziptie.dev/blog/how-ai-remembers-your-content-across-sessions/) 3. [Replika Review 2026: The AI Companion That Got Fined €5M - Scribe](https://scribehow.com/page/Replika_Review_2026_The_AI_Companion_That_Got_Fined_euro5M__Is_It_Still_Worth_Using__siVxaR7TShSgBhaRaZ6osg) 4. [How to Make a Character on Character AI (2026 Guide) - Ai Insights](https://aiinsightsnews.net/how-to-make-a-character-on-character-ai/) 5. [Replika vs Nomi (2026): Finding Enduring AI Companionship](https://nomi.ai/ai-today/replika-vs-nomi-2026-finding-enduring-ai-companionship/) 6. [Pi AI Review 2026: Your Always-On AI Companion | Agent Finder](https://agent-finder.co/reviews/pi-ai)