Beta on TestFlight, iPhone only

Your AI assistant runs on your iPhone.Turn on airplane mode and keep talking.

Elren plans your day, changes your calendar and remembers what you tell it. The model runs on your phone's chip, so you can use it with no signal and no token limit.

Join the betaSee how it works

A chatbot answers questions.Elren also does the work in your apps.

Works with no signal

Use Elren on a plane, in the subway or abroad with roaming off. The model, your memory and your calendar all live on the phone.

Chatoffline
Memoryoffline
Todosoffline
Calendaroffline
Remindersoffline
Files and scansoffline
Gmailneeds signal
Web searchneeds signal

Gets better the more you use it

Correct Elren once and the next reply uses the correction. Turn on cloud learning and it also trains a personal layer on your examples, from once a week up to once a day.

I run in the evenings now, not mornings.
Updated. I'll plan your runs after 18:00.
memory replaced: "runs in the morning"

Your tokens never run out

Cloud assistants bill by the token, so a limit can stop you mid-task. Elren uses your phone's chip, and you can send as many messages as your battery allows.

Tokens per monthno limit
Messages per dayno limit
Fee per message$0

Web search calls a paid service, so it has a cap of 100 searches a day.

Connected to the apps you use

Ask for an outcome and Elren uses the right app. It marks a task done only after that app confirms it, and it sends email only after you approve the exact message.

CalendarGmailFilesRemindersTodosWeb
Email to Sam, waiting for youRe: Thursday's review

Thursday at 10:00 works for me. I'll bring the draft.

Your phone does the thinking.Servers handle sign-in, plans, search and opt-in training.

Your iPhoneruns offline
Chat and approval cardsCards show what your apps returned, never the model's wording.
Agent loop and context builderRebuilds the prompt from your memory, todos and calendar before each model run.
Local modelA 0.8B model reads each request. It asks before loading 2B, 4B or 9B, on llama.cpp and the phone's GPU.
Personal layerAn optional adapter trained on your examples.
Encrypted memorySQLCipher database with full text and vector search.
Tool executorValidates each call, runs it once and records the result.
Calendar and remindersReads and writes through iOS.
Files and scansApple Vision reads screenshots and PDFs on the phone.
Over the internetneeds signal

Google

GmailYour phone calls Google itself. Mail skips Elren's servers.

Elren servers

Sign-in and plansStores an Apple ID reference and your plan state.
Search proxyHides the search key and applies the daily cap.

Cloud learning, off by default

Disposable GPUOne pod per job. It trains, reports and shuts down.

Flip the switch to see what keeps running.

01

Context

Elren collects the memories, events and todos that fit your request.

02

Decision

The model on your phone picks a tool or writes a reply. Text from mail and web pages reaches it marked as untrusted.

03

Checks

Code validates the call, compares dates with weekdays and blocks repeats. A cut-off call never runs.

04

Result

iOS or Gmail returns what happened, and that result becomes the card you see.

Your memory updates right away.Your personal model updates on a schedule.

Memory stores facts. The personal layer teaches the model how you like things done. You decide if it runs, and you can reset it at any time without losing your memories.

1. Opt inYou chooseCloud learning stays off until you turn it on. Your phone prepares examples from your chats and corrections, and you send them with a tap.
2. UploadEncrypted storageYour phone sends the examples over TLS. The server stores them with AES-256 and deletes them after 7 days.
3. TrainA pod just for youA temporary GPU pod trains a small adapter on your examples, and Elren destroys the pod after the job.
4. TestA test before releaseThe adapter must beat the base model on examples it never saw and pass a behavior test. Otherwise Elren discards it.
5. InstallBack on your phoneYour phone checks the file's hash and loads the adapter. You can roll back to the previous version.

If you turn learning off, Elren cancels any running job and deletes your uploaded examples.

Pay for a bigger model and more learning.Every plan has unlimited local messages.

No plan

$0

  • Compact model
  • Memory, calendar, mail and files
  • No cloud learning

Basic

$4.99 / month

  • Compact model
  • Learning once a week

Pro

$9.99 / month

  • Compact and Large models
  • Compact learns every 3 days
  • Large learns once a week

Premium

$19.99 / month

  • Compact and Large models
  • Both learn once a day

Beta prices. They may change before public release.

Questions

Which iPhones can run Elren?

Elren starts with a 0.5 GB model for reminders, todos and events, and that needs about 3 GB of memory. For harder requests it offers a bigger model and you tap to load it. Light needs about 4 GB of memory, Compact about 6 GB and Large about 12 GB. The app checks your phone before it downloads anything.

What leaves my phone?

Web searches go through Elren's proxy. Gmail requests go to Google. Sign-in and plan checks go to Elren's server, and the app downloads its models once from Hugging Face. Your training examples leave only if you turn on cloud learning.

Does it match the largest cloud models?

No. Elren runs models with 0.8 to 9 billion parameters, small enough for a phone. Use it for your schedule, tasks, mail and notes, and use a large cloud model for deep research.

Will it drain my battery?

Long sessions use more battery and warm the phone, because the model runs on the phone's GPU. Short requests cost little.

How do I delete my data?

Settings has one action that wipes the phone's local data and another that deletes your account and everything on the server. You can also view, correct or forget single memories in the Memory tab.

Elren is in beta on TestFlight.Send us a note and we'll add you.

Join the beta