Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Mobile & Edge LLMs

PocketLLM Lite Review: Local-First AI Chat App for Android

·6 min readBy Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

PocketLLM Lite (version 1.0.40, released on 15 October 2026) is a free, open-source Android chat app that runs models on the phone or on a server you control. On-device inference loads only Cactus-format model bundles, not GGUF files, and the project labels its APKs as debug-signed engineering builds.

Key Takeaways

  • Maintenance: the repository was last pushed on 15 October 2026 and several releases have followed since August, so no stale-project warning applies.
  • Cost: free, with no account, and per the README no ads or analytics SDKs.
  • Runtimes: on-device Cactus bundles, an Ollama server or any OpenAI-compatible endpoint.
  • Signing: release APKs are debug-signed, and the maintainers call them engineering builds.
  • Maker: no named developer; the project is run by a GitHub organisation.

📍 In One Sentence

PocketLLM Lite is an MIT-licensed Android app that chats with models running on the phone or on your own Ollama or OpenAI-compatible server.

💬 In Plain Terms

You install an APK from GitHub, download a small model into the app or point it at a server at home, and then chat or ask questions about your own documents without an account.

📌Note: This review is based on the project's website, GitHub release v1.0.40 and repository documents, checked on 24 October 2026. PromptQuorum has not installed or tested the app, so speed, answer quality and stability are not assessed.

Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

PocketLLM Lite is a free, MIT-licensed Android chat app that runs compatible models on the phone or connects to your own Ollama or OpenAI-compatible server. No developer is named; the code lives in the PocketLLM GitHub organisation and the product has its own site at pocketllm-lite.vercel.app. This review is based on that site, GitHub release v1.0.40 and the repository documents, checked on 24 October 2026; PromptQuorum has not installed or tested the app.

PocketLLM Lite Review: Local-First AI Chat App for Android

What Is PocketLLM Lite?

PocketLLM Lite is a local-first AI workspace for Android, built with Flutter, with a companion web edition. The project describes it as an app where chats, documents, memories and models stay under the user's control; the GitHub repository was created on 9 December 2025.

It separates the app from the inference backend and offers three routes: on-device models through the Cactus runtime, an Ollama server on loopback or a trusted local network, and any configurable OpenAI-compatible endpoint.

Get It

PocketLLM Lite is distributed as APK files on GitHub Releases, plus a browser edition; no store listing is linked and no iOS app is offered.

Option
How to get it
Android, most phonesarm64-v8a APK, about 60 MB, includes the on-device runtime
Android, older or x86armeabi-v7a and x86_64 APKs; servers, OCR and documents only
Web editionOpen the web app; needs your own Ollama or OpenAI-compatible endpoint
Websitepocketllm-lite.vercel.app
Source codeGitHub repository, MIT license

Version reviewed: 1.0.40, published on 15 October 2026 with a SHA256SUMS file; Android 7.0 (API 24) or newer is required. The APKs are debug-signed, so Android shows no verified publisher, and installing needs permission for unknown sources. According to the README, a 1B-parameter model at Q4 takes roughly 0.7 to 1 GB.

How to Get Started

The steps follow the project's own five-minute quick start; PromptQuorum has not run them.

  1. 1
    Install the APK
    Why it matters: Download the arm64-v8a file from the release page, compare it with SHA256SUMS.txt and install it.
  2. 2
    Finish onboarding
    Why it matters: Choose your network permissions; update checks and online model browsing are optional.
  3. 3
    Pick a runtime
    Why it matters: Download a small catalog model for on-device use, or enter the URL of an Ollama or OpenAI-compatible server in settings.
  4. 4
    Chat or import documents
    Why it matters: Per the README, the selected model produces every answer, and the app says so when a prerequisite is missing instead of faking a reply.

Features Confirmed by the Sources

Every item below comes from the README or the release notes; none has been independently tested.

  • Chat workspace. Streaming chat with history, branches, tags, personas, prompts, skills and per-chat system prompts.
  • Documents and memory. PDF, TXT, Markdown and CSV import with source references; semantic search needs an installed embedding model, and local memories can be inspected, pinned and disabled.
  • Voice and OCR. On-device Whisper transcription after a checksum-verified model install, text-to-speech, and Android OCR for Latin text.
  • Tools. Validated JSON tool calls such as math, notes, reminders, clipboard and policy-gated web search; state-changing tools ask first.
  • Developer features. Prompt Lab, side-by-side model comparison, on-device benchmarks and an authenticated OpenAI-compatible server inside the app, bound to loopback by default.

Data Flow and Privacy

The README says chats, documents, indexes and memories stay in the app sandbox, on-device inference sends nothing out, and no ads, analytics or tracking SDKs are included. A regression test in the repository is said to fail the build if such SDKs appear.

Network use is meant to be explicit: model discovery and downloads, update checks, web search through Tavily with your own key and remote runtimes you configure contact their displayed destinations. Strict Offline Mode blocks non-loopback traffic from the app, an audit log lists destinations, and backups use AES-256-GCM with a password-derived key. A prompt sent to an Ollama or OpenAI-compatible server leaves the phone for that server.

📌Note: Strict Offline is an application-level boundary, not a device firewall, and the project's security notes say the optional in-app API server has no TLS and no CORS allowlist, so it belongs on loopback or a trusted network. These are the project's own statements; PromptQuorum has not inspected network traffic.

Who Should Use It

  • Android users who already run Ollama and want a phone client with document search and an audit log.
  • Tinkerers who want an MIT-licensed codebase they can read, build with Flutter and adapt.
  • People curious about on-device models who accept sideloaded, debug-signed builds and an unproven runtime.

What We Could Not Verify

  • On-device inference. The release notes say token generation was not verified on a physical phone and model loading only on an emulator, so speed, memory fit and stability are unknown.
  • Site versus repository. The website lists "On-device GGUF", while the README says GGUF files cannot run on the phone; the repository description also mentions iOS although the README covers Android only. This review follows the README.
  • Signing and publisher. Debug-signed APKs give no publisher identity, and no developer is named.
  • Runtime upkeep. The known-limitations file says the Cactus Flutter package is archived upstream as of July 2026 and lists that as a maintenance risk.
  • Privacy and quality. The no-tracking claim, the audit log and backup encryption were not inspected, and answer quality was not assessed.

Competitors and Alternatives

Maid

Platforms:
Android, iOS
Price / license:
Free / MIT
Key difference:
Open-source Flutter app that also works with Ollama and OpenAI-compatible APIs
Articles about Maid (5)

Also mentioned in:

These are the nearest mobile chat reviews on PromptQuorum; confirm each app's current price, license and platforms on its own page.

Compare in directory

Frequently Asked Questions

What does the web edition do?

It is a Next.js browser version that keeps its data in IndexedDB and talks to an Ollama or OpenAI-compatible endpoint you control; the README says it does not include the Android on-device runtime.

Which APK should I download?

Use arm64-v8a for most phones from 2017 onward. The armeabi-v7a and x86_64 builds cannot run on-device models because the runtime library is packaged for arm64-v8a only.

Does it work without internet?

The project says on-device chat needs no network once a model is installed, and Strict Offline Mode can block the app's own non-loopback traffic; the first model download and web search do need a connection.

Is the app available in other languages?

No. The README says the interface is English only today; translation files for six languages exist but hold 15 strings and are not wired in.

Verdict

PocketLLM Lite is an actively updated, MIT-licensed way to put a documented local-first chat and document workspace on an Android phone, with a clear choice between on-device models and your own server. Against that, it has no named developer, ships debug-signed engineering builds, depends on an upstream runtime package it lists as archived, and its authors say on-device generation has not been verified on a real phone. It suits technical Android users who can sideload, read the repository and test the app themselves; anyone who needs a signed, store-distributed app with verified performance should wait for those to appear.

Sources

← Back to Power Local LLM

MCP