Documents & RAG
Ingest PDF, TXT, Markdown and CSV files. Use lexical, semantic or hybrid retrieval with source-aware context.
Your Personal
AI, on your terms.
PocketLLM Lite is a local-first AI workspace built around a simple rule: your conversations, documents, memories and models should stay under your control.
Android
Local models
Doc RAG
Ollama
Latest release
PocketLLM Lite
It’s not another AI account.
It’s your own AI workspace.
No cloud profile is required to chat, save history, use personas, build local memory, or work with documents. Network features are explicit, inspectable and optional.
Press Enter to continue in the web app — your prompt carries over.
See how local-first works ↗Own your data
The app is yours,
the data is yours.
Chats, local memories, prompts, personas, skills and document indexes stay in the app sandbox. Strict Offline can block non-loopback app-managed network traffic before a request leaves the app.
Private by defaultNo analytics or advertising SDK required for the core experience.
Inspect the boundaryNetwork destinations and purposes are visible in the audit log.
Back it upPassword-derived AES-256-GCM backups are portable and verifiable.
Local Assistant · Sandbox Active
Did you know PocketLLM runs completely offline?
Yes! All model weights run directly on your phone’s NPU/CPU. Chats, document indexes, and private memories remain strictly inside your device sandbox.
Explore
Tools for your thinking.
Not a single giant magic button. A proper local workspace with composable pieces you can understand and control.
Personas & skills
Build reusable assistants, prompts and skills. PocketLLM composes them into the generation pipeline instead of hiding the prompt stack.
Memory, locally
Inspect, pin, disable and supersede local memories. Retrieval falls back honestly when embeddings are unavailable.
Confirmed tools
Calculator, notes, reminders, clipboard, links and web search use schema validation and visible confirmation where required.
Voice & audio
Dictate, listen and transcribe through supported local or platform paths, with capabilities shown instead of guessed.
Honest model setup
Browse, import and manage models with explicit source, size, license and compatibility state. Unknown stays unknown.
Choose your route
One workspace.
More than one runtime.
PocketLLM separates the product from the inference backend. Pick the route that makes sense for your device and your privacy boundary.
01On-device GGUFCompatible local models through the supported mobile runtime.
02OllamaUse a loopback or trusted LAN Ollama endpoint you control.
03OpenAI-compatibleConnect a configurable endpoint only when you want remote inference.
Download
Get PocketLLM Lite.
Android builds are published on GitHub with release notes and checksums. Choose the universal APK if you are unsure about your device ABI.
Current GitHub release
PocketLLM Lite
Loading release information…
For builders
Build it for yourself,
see it run everywhere.
PocketLLM Lite is MIT licensed. Read the source, reproduce the behavior, inspect the tests and contribute focused improvements. The project prefers evidence over marketing folklore, which is refreshing in AI land.
Local AI, properly
A smaller app with
clearer boundaries.
PocketLLM is being built around inspectable capabilities, portable data and runtime choice instead of tying your AI workspace to one vendor or one account.
Read the docs