PocketPal AI and Merrin both run a language model on your phone and keep your conversations on the device. That is where the overlap ends.

This is PocketPal AI vs Merrin as we see it, written by the team that builds Merrin, so assume we are biased about the conclusion and careful about the facts. PocketPal starts with a question about the runtime: which model do you want to run? Merrin starts with a question about you: what do you want to talk about? One is a control panel. The other is a personal intelligence that remembers, and that difference is what most of this article is about.

What Merrin is like to use

Merrin begins with a blank conversation and no setup. There is no model picker, no quantization list, no benchmark screen. The app manages the model so your attention stays on the conversation, not the runtime.

  • Merrin's Today screen: "Good evening. What's on your mind?", the Tell Merrin anything field with a voice button, "3 chats today", Pick up cards for "Should I take the new role?" and a plan still open, and earlier chats titled Training plan for the 10K and Call Leah on Sunday
  • Merrin's You tab: the Life right now card with the plans and people from recent chats (Finish the 10K under an hour; Leah, sister; Book the pottery class; Claire, new manager; Settle into the new job), above tiles for 4 people and 18 memories
  • Merrin's Insights on the Now tab: the week of Nov 2–8 with 25 entries and the line "A new job and a 10K to train for", above active threads sized by entries
Real screens from Merrin: Today, where you start, the people and plans Merrin remembers, and Insights (the full weekly reflection is part of Pro).

It remembers what matters, and you control what it keeps

Merrin builds a memory of the things you would want an assistant to know: the people in your life, a trip you took, a goal you mentioned, a decision you made. Memories are typed, linked back to the message that created them, and editable. You can inspect, correct, or delete anything it keeps from the memory controls on the home page. Memory is on the free plan; Pro adds a longer view across months. That is the difference between an archive and a memory. A chat log stores every message and understands none of them. A memory system decides what is worth carrying forward, and it keeps the receipt.

You can ask about something you said weeks ago

Because memories and journal entries are indexed by meaning rather than by text, you can ask in the words you have today: “What was that ramen place I liked in Tokyo?”, or “Wasn’t I going to catch up with Alex?”. Merrin finds the Ichiran note from your Japan trip and the friend you mentioned wanting to see, even though neither conversation used those exact words. Semantic search runs across conversations, journal entries, and memories together.

Your conversations become a journal you keep

Conversations become journal entries automatically, so a week of thinking does not disappear into a chat list. Entries live in the same searchable history as the conversations they came from, and with Pro, the weekly reflection turns what you wrote into something short you can keep.

It notices the patterns you would miss

Insights look across months, not messages. Merrin surfaces themes like goals, events, decisions, and worries, and shows how they change over time. A useful one is specific: “You have mentioned changing direction on this project four times in the last month”. That is a pattern you can act on, not a personality guess. Active threads are free; seeing how they change over time, and weekly reflections, are part of Pro.

It keeps the small promises

Merrin extracts actions from conversation, so “remind me to message Alex on Friday” becomes a reminder attached to the person and the date, not a sentence you have to find again. Small, but it is the difference between an AI that talks and one that helps.

Private by architecture, honest about the limits

Conversations, journal entries, memories, embeddings, and insights are designed to stay on your device during normal use. Model downloads and purchases are the limited network calls, and the privacy policy describes exactly where the boundary sits. Two honest limits: on-device models are smaller than frontier cloud models, so they are weaker on general knowledge questions, and a local journal is only as safe as your device, so keep your own backup. We wrote a policy review of the cloud assistants if you want to compare where else a journal could live.

The loop that makes it feel personal

Here is the part that is hard to see in a feature list. Memory is not one feature; it is a loop: conversation, extraction, memory, retrieval, context, journal, search, insights, and the next conversation. Each stage makes the next one better, and the loop compounds the longer you use it. None of it is a black box, either: every memory in the loop is something you can see, edit, and delete, which is what makes a compounding loop safe to trust.

flowchart LR
  C[Conversation] --> X[Extraction]
  X --> M[(Memory)]
  M --> R[Retrieval]
  R --> K[Context]
  K --> C
  X --> J[Journal]
  J --> S[Search]
  S --> C
  M --> I[Insights]
  I --> C
The personal continuity loop. Every stage feeds the next conversation, which feeds every stage again.

Tomorrow’s conversation starts where today’s ended. Next month’s search finds what last month’s entry actually said. That is why we spend our engineering time on the loop instead of on a longer list of models. The engineering deep dive explains how the memory loop works under the hood.

What it feels like after a month

  • Day one. You talk about your week. Merrin asks a follow-up question or two. Nothing magical yet; it feels like a good conversation.
  • Week one. You mention a friend, a project, a worry. When you come back, the app knows what you meant without a recap.
  • Month one. You ask something you would normally search your notes for, and the answer includes the entry it came from. You notice a pattern in your own history that you would not have seen on your own.

That compounding is the product. It is also why the choice between these two apps is not really about models. PocketPal gives you a model to run. Merrin gives you a memory that grows.

PocketPal AI vs Merrin: which should you choose?

The wider trade underneath this comparison — local model control versus a managed, memory-first experience — is the same on-device versus cloud question we scored axis by axis. Narrowed to the two apps:

DimensionPocketPal AIMerrin
Primary userPeople who enjoy running and tuning modelsPeople who want a private personal intelligence
Core experienceChoose, run, and configure modelsTalk, journal, and reflect
ModelsThousands of GGUF models, you chooseManaged for you
SetupPick a model and size it to your phoneOpen the app and talk
OfflineYes, after a model downloadYes, after the required models are installed
VoiceOn-device text-to-speech, an optionVoice-first; voice is free
MemoryChat history and pinned chatsMemory system with inspect, edit, delete; Pro adds a longer view
JournalNoCore feature; conversations become entries automatically
SearchWithin chat historySemantic search across conversations, entries, and memories
InsightsNoActive threads; Pro adds view changes and weekly reflections
ActionsNoExtracted from conversations, such as reminders
Model controlsExtensiveHidden by design
PriceFree and open source (MIT)Free, with optional Pro at $29.99 per year or $99.99 once
Scores are our read of each product's shipped priorities, not benchmark results. PocketPal leads on model control, by design. Merrin leads on the continuity axes. Choose the shape that matches what you want.

Model control points to PocketPal. Continuity points to Merrin. Keeping both is fine too: PocketPal for curiosity about models, Merrin for the parts of your life you want carried forward.

A private place for what you don’t want to forget

If you want to run any model, PocketPal is a good home for that. If you want an AI that remembers what you talked about last month, Merrin is free to download, and you can see how the memory controls work on the home page. If you are building in this category and want to compare notes, talk to us. We have made most of these mistakes already.

Frequently asked questions

What is the difference between PocketPal AI and Merrin?

PocketPal AI is a local model runtime: you choose and configure the model, and the app's focus is inference. Merrin is a private personal intelligence: the app manages the model, and its focus is continuity, meaning memory, journal entries, semantic search, actions, and insights over time.

Does Merrin work offline?

Yes. After the required models are installed, core use works without a connection. Model downloads and purchases are the limited network calls, and neither carries the content of your conversations.

Can I edit or delete what Merrin remembers?

Yes. Memories are linked to the messages that created them, and you can inspect, edit, or delete them from the memory controls. The controls apply to everything the app keeps, on every plan.

Is Merrin free?

Merrin is free to download on the App Store and Google Play. The free plan has no message limits. Voice, automatic journal entries, memory, and search by meaning are included. Pro is optional, at $29.99 per year or $99.99 once, and adds insights over time, weekly reflections, more perspectives, and a larger model.

Can I just keep a long chat history instead of using memory?

You can, and most chat apps do. A long history is a scrollback: it stores everything, understands nothing on its own, and the model still starts each conversation from scratch. Memory is the work of deciding what matters, connecting it to the rest of your history, and bringing it back at the right moment. That work is the product.

Is PocketPal AI private?

Yes. PocketPal runs inference on your device, has no account requirement, and is open source, so its privacy claims can be checked. Both apps are private in the same way; the difference is what each one does with your history.