# Pastura - Local LLMs simulator — Like stargazing, but for LLMs

> Like stargazing, but for LLMs (iPhone/iPad app by TOMOHITO YABU.)

- Source: https://appshunter.io/ios/app/pastura-local-llms-simulator/id6788409688 (this page in markdown: same URL + `.md`)
- Developer: [TOMOHITO YABU](https://appshunter.io/developer/6779960632)
- Category: Developer Tools, Entertainment
- Price: Free
- Age rating: 12+
- Requires: iOS 18.0 · 17 MB
- Languages: American English, Japanese
- Released: 2026-07-23
- Data updated: 2026-09-07
- User reviews in markdown: https://appshunter.io/ios/app/pastura-local-llms-simulator/id6788409688/reviews.md

## What is Pastura?

AIgazing. Like stargazing, but for local LLMs.

Pastura is a closed pasture for AI agents, running entirely on your iPhone. You write a scenario. The agents act it out. You step back and watch their speech, their inner voice, their votes and scores arrive in real time.

You can't chat with them, and that's the point. Once you join in, the agents start reacting to you, and the natural exchange between them disappears. Pastura isn't a chat app for talking to an LLM. It's an app for watching LLM agents think.

Everything runs on-device. No account, no sign-up, and none of your data ever leaves your phone. Airplane mode is fine.

WHAT YOU CAN DO
• Run built-in scenarios like Word Wolf or Prisoner's Dilemma
• Write your own in a visual editor, or switch to YAML for full control
• Browse a curated gallery of shared scenarios and import with one tap
• Swap between local LLMs, and see how the same scenario changes
• Watch speech, inner voice, votes, and scores arrive live
• Export any run as Markdown to keep, paste, or share

WHY ON-DEVICE
The LLM lives on your phone, not on someone else's server. That settles privacy, cost, and latency in one move. Inference never touches a server. Zero monthly bill. Zero telemetry, zero analytics. Your scenarios, transcripts, and results never leave the device.

The first time you pick a model, Pastura downloads it once (around 3 GB, Wi-Fi recommended). After that, it's yours and it's offline. While you wait, a built-in demo run plays so you can see what observation looks like before the model finishes downloading.

A BIGGER PICTURE
Stargazing taught us to look up at the stars. AIgazing teaches us to look at the agent. We rarely stop to observe our AI. Pastura is a quiet window into that, on-device, yours alone. Whether it tells you something about them, or about you, is left as a question for the observer.

Requires a device with 6.5 GB of RAM or more (iPhone 15 Pro and newer). Free, with no subscription and no in-app purchases.


## Version history (last 3 releases)

### 1.2 — 2026-08-28

Dark mode
- Pastura now supports dark mode, and switches with your device setting

Models
- New recommended model: Gemma 4 E2B (QAT), a lighter download than before
- A model you already downloaded stays where it is and keeps working

Browse
- Scenario pages now show excerpts from a real run, laid out like the simulation
  screen (six scenarios so far)
- The list stays where you left it when you come back from a scenario
- Scenarios you already added are grouped at the end

Stability
- Improved the reliability of AI responses
- Other minor fixes

### 1.1 — 2026-08-04

New preset scenarios
- Added "The Inheritance Meeting": each heir argues a public case while pursuing a hidden agenda
- Prisoner's Dilemma gained a whisper phase, a round-progress display, and an endgame rule
- Word Wolf now has live commentary

More expressive simulations
- Commentary: a narrator calls out the highlight of each round
- Hidden agendas: a character can hold a private goal behind what they say in public

Sharing
- Share a single line as an image card, with the character's inner thoughts included if you want
- When a simulation ends, lines worth sharing are suggested for you
- Share straight to X and Instagram Stories
- Opening a shared link now launches the app

Editor
- Set a sentence-length limit for each phase
- Clearer labels and inline descriptions for variables and fields

Stability
- A turn the AI fails to answer is now marked as skipped and kept in the result
- An interrupted simulation can resume where it stopped
- Vote ties resolve consistently, and dialogue repeats itself less often
- Fixed a display glitch on some screens in dark mode
- Other minor fixes

### 1.0 — 2026-07-23

No release notes.

## Related topics

[local llm server](https://appshunter.io/ios/topics/local-llm-server)

---

*Data collected daily from the US App Store and indexed by [AppsHunter](https://appshunter.io/). User reviews are verbatim App Store reviews. Ratings, prices and chart positions refresh continuously; this snapshot is from 2026-09-07.*
