Model or dataset
Vali-98/ChatterUI avatar
Vali-98/ChatterUI

ChatterUI ships an APK, skips iOS, and carries its version number two minors behind

A frontend for running models on mobile or connecting to your preferred API providers.

2,910 stars279 forksTypeScriptAGPL-3.0

At a glance

What is it?
A React Native and Expo frontend for llama.cpp models and remote APIs whose package.json still reads 0.8.8 while the newest tag is v0.10.0, and whose install step patches four vendor packages.
Who is it for?
ChatterUI is a frontend with two backends behind it and no server of its own, which makes it a good fit for someone who wants character cards, sampler control and instruct formatting over a model they already run. It is not a fit if you need an iOS build today or a store listing, because the README states iOS is unavailable for lack of development hardware and the only install path is an APK from the releases page.
Can I use it commercially?
Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
Is it still maintained?
Yes. The repository last received commits 2 days ago.
What is it written in?
Mainly TypeScript, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • Vali-98/ChatterUI2,910 ★ · +116
Star history
Repository2026-10-042026-11-15
Vali-98/ChatterUI2.8k2.9k

116 Stars in the last 42 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

iOS is unavailable, and the iOS scripts are still there

The distribution story is one sentence: download and install the latest APK from the releases page. Underneath it sits an italic line saying iOS is currently unavailable due to lacking iOS hardware for development. That is a candid reason and an unusual one to find in a README.

The build tooling does not agree with it. package.json carries an ios script running expo run:ios and a dev:ios variant that sets APP_VARIANT to development before doing the same, alongside the matching android and dev:android scripts and a web script that starts the bundler for the browser. The repository root holds app.config.js, metro.config.js, babel.config.js and an eas.json.example, which is the Expo Application Services configuration kept as an example rather than committed directly.

So the project is configured for three platforms and distributed on one. For anyone planning a release, that gap decides the roadmap: an Android sideload is available today, and an iOS build is a porting job that the maintainer has explicitly not done yet.

package.json reads 0.8.8 while the newest tag is v0.10.0

The version metadata has drifted from the releases. The version field in package.json is 0.8.8. The three newest tags are v0.10.0 published on 2026-09-22, v0.10.0-beta5 on 2026-08-28 and v0.10.0-beta4 on 2026-08-19, so the repository is two minor versions ahead of the file that npm would read if the package were published from a checkout.

The release pattern underneath is a long beta cycle for one minor version: two betas nine days apart, then the final tag about a month after the first. The last push to master is dated 2026-09-22, the same day as v0.10.0 and several hours after it, which means the tag and the branch head are close but not identical.

Nothing else in the file compensates for this. main points at expo-router/entry, so the package is an application rather than a library, and there is no private flag or publishConfig to read. The practical effect is that the version in the file tells you nothing about which build you installed.

Local Mode is llama.cpp on gguf files, and the file has to fit in memory

On-device inference runs through llama.cpp, with a custom adapter called cui-llama.rn written to bridge it to React Native. The acknowledgements credit llama.cpp as the underlying engine and llama.rn as the original React Native adapter, so the fork relationship is acknowledged rather than hidden.

Two paths exist for getting a model in, and the trade-off is storage against startup time. Import Model copies the gguf file into the app, which can speed up startup. Use External Model points at a file already on device storage, which avoids copying large files into the app at the cost of a slight delay in load times. The first choice in Local Mode is Models, then Import Model or Use External Model, and the constraint is stated plainly: pick a gguf model that can fit on your device's memory.

There is one hardware specific recommendation. For devices with Snapdragon 8 Gen 1 and above, or Exynos 2200 and above, Q4_0 quantization is recommended for performance. No other quantization guidance is given for other chipsets.

Remote Mode has three backend tiers and a template system

The remote side is organised in tiers, which is more structured than most clients of this kind. Open source backends are named first: koboldcpp, text-generation-webui and Ollama. Dedicated APIs come next: OpenAI, Claude, Cohere, Open Router, Mancer and AI Horde. Generic backends are the third tier, Generic Text Completions and Generic Chat Completions, with a note that these should be compliant with any text completion or chat completion backend such as Groq or Infermatic.

That third tier is where the hedge sits. The wording is that the generic endpoints should be compliant with backends of that shape, which is a statement about the request format rather than a tested list of providers. Nothing in the README records which services were verified.

The fourth path is a template system for providers the app does not ship a preset for, and it is documented in a discussion rather than in the README. So the answer to a missing provider is to write a template rather than to wait for a release.

Install runs patch-package against three vendor packages

The install step is not a plain dependency resolution. postinstall runs patch-package, and three dedicated scripts exist to build those patches:

json
"postinstall": "patch-package",
"migrate": "npx drizzle-kit generate",
"lint": "eslint app db lib --ignore-pattern db/migrations"

patchsql patches expo-sqlite, excluding the .project, .cxx, .c, .h, build and bin paths and turning on case-sensitive path filtering. patchfetch patches expo itself, limited to fetch.ts files and NativeResponse.kt. patchfs patches expo-file-system for .kt files. The patches live in their own directory at the root, and the same patch-package tool that consumes them runs on every install.

This is the part of the project with real upgrade risk. An upstream release that renames a file or shifts a line inside those three packages turns the patches from invisible plumbing into a failed install, and the fix has to land in this repository before your build works again.

Character cards, sampler fields and a drizzle-managed database

The chat layer is where the app differentiates itself, and the README describes it as fine-grained control over chat structuring. Characters use the Character Card v2 specification, and each character can hold multiple separate chats. Beyond that, sampler fields and instruct formatting are both exposed for customisation, and the app integrates with the device's text-to-speech engine rather than shipping audio of its own.

State lives in SQLite through drizzle. The root carries drizzle.config.ts, a db directory and an eslint ignore pattern for db/migrations, while the migrate script generates those migrations with drizzle-kit generate. Tests run through jest with the jest-expo preset, and the jest configuration carries a moduleNameMapper that redirects expo-modules-core to the copy nested inside node_modules/expo, which is a second sign that the Expo dependency tree needed help.

An i18n directory sits alongside app, lib and assets, so the interface is set up for translation even though the README does not list the languages.

One APK, one funding file, and no store listing

The project is funded in two small ways and distributed in one. FUNDING.yml sits at the root of the tree, and the README carries a ko-fi link for the maintainer. There is no homepage field recorded for the repository.

Distribution is a single sentence pointing at the latest release APK, with no mention of a store listing of any kind. For an Android user that means enabling installation from unknown sources and tracking releases yourself; for anyone who needs an update channel, a store or an MDM, the project does not offer one in writing.

The development documentation is also thinner than the rest of the README. It points at CONTRIBUTING.md under docs, and the only other material in the repository is the discussions area, which is where the custom API template system is explained. So the code is readable and the extension point is documented, but the onboarding path is a single link.

Editorial conclusion

ChatterUI is a frontend with two backends behind it and no server of its own, which makes it a good fit for someone who wants character cards, sampler control and instruct formatting over a model they already run. It is not a fit if you need an iOS build today or a store listing, because the README states iOS is unavailable for lack of development hardware and the only install path is an APK from the releases page. Before you build on it, check three things. Read the version in package.json rather than the tag, since the file says 0.8.8 while the newest release is v0.10.0. Plan for the patched dependencies, because postinstall runs patch-package against expo, expo-sqlite and expo-file-system, so an upstream change can break your install outright. And decide how you will get the model onto the device, since copying a gguf file into the app speeds up startup at the cost of storage, while pointing at external storage saves the copy and adds load time.

Frequently asked questions

What is ChatterUI?

It is a native mobile frontend for LLMs built with React Native and Expo, licensed AGPL-3.0. In Local Mode it runs gguf files on device through llama.cpp with a custom adapter, and in Remote Mode it connects to open source servers, dedicated APIs or generic completion endpoints.

How to install ChatterUI on Android?

Download the latest APK from the releases page and install it on the device. On-device inference then needs Local Mode enabled, followed by Models, Import Model or Use External Model, choosing a gguf file that fits in the device's memory. iOS is stated as currently unavailable for lack of development hardware.

What is a chat UI?

The repository never defines the term, using it only as part of the project's own name. What it does describe is a mobile frontend with character cards, multiple chats per character, customisable sampler fields and instruct formatting, and integration with the device's text-to-speech engine.

What is the best open-source chat app?

The repository makes no comparison and names no competitor in this category. It credits llama.cpp as the engine underneath and llama.rn as the original react-native adapter, ships its own adapter named cui-llama.rn, and points at its own releases page for the APK.

How does ChatterUI compare to PocketPal?

The repository contains no comparison with PocketPal and does not mention it. What it records is its own backend list: koboldcpp, text-generation-webui and Ollama for open source servers, OpenAI, Claude, Cohere, Open Router, Mancer and AI Horde as dedicated APIs, plus generic text and chat completion endpoints and a template system for anything else.

Does ChatterUI have an iOS version?

No. The README states iOS is currently unavailable due to lacking iOS hardware for development, and the only install path it gives is the latest APK from the releases page. The package scripts still include expo run:ios and a dev:ios variant that sets APP_VARIANT to development.

Official sources

  1. Issues
  2. License: AGPL-3.0
  3. README
  4. Releases
  5. Vali-98/ChatterUI on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/vali-98-chatterui.svg)](https://hysenlabs.com/projects/vali-98-chatterui)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/vali-98-chatterui.svg)](https://hysenlabs.com/projects/vali-98-chatterui)

Was this useful?