LatentChat - Assistant LLM

Try different generative language models with LatentChat, a fully offline, on-device private AI chatbot.

LatentChat is ready to use right out of the box. Ask questions, brainstorm, reason, transcribe speech, add your own models and create with your customized chat companion from anywhere, even without an internet connection.


- Local chat
Try models like Llama, Gemma, Qwen and Mistral or use your own local LLMs by selecting your preferred open-source language models stored on your device. Reasoning models are supported, too.

- Remote chat
LatentChat can be used as a client to connect to any OpenAI-compatible and Anthropic-compatible API endpoint, whether it is running on your own private servers like Ollama, LM Studio or on cloud services OpenRouter (Bring your own API Key).

- Retrieval-Augmented Generation (RAG)
Create your knowledge center from text or files (PDF, TXT, JSON, HTML) and retrieve your documents while you chat to ask related questions or summarize.
A RAG system can also help the model reduce hallucinations by referencing the knowledge you added.

- Vision capabilities
Try multimodal models like Qwen 3 VL, InternVL and LFM2.5 with vision capabilities, to analyze images or extract text.

- Transcription and Voice (STT and TTS)
Transcribe recordings or audio files with Whisper and read messages aloud with different voice styles. Multiple languages are supported.

- System Prompts
Use System Prompts to personalize your conversations by providing custom instructions that can modify the behavior or personality of the AI.


Features:
== Completely offline and on-device processing
== Your prompts remain private and are never shared
== Model catalog: a selection of LLMs downloadable from HuggingFace
== Bring your AI models: download and choose your favorite LLMs (.gguf)
== Connect to servers like Ollama or LM Studio for more powerful models
== Local RAG: Integrate your own knowledge from text
== Vision capabilities with multimodal models to process images
== Text to speech: read messages aloud (English, Spanish, French, Portuguese)
== Multilingual transcription (STT) using Whisper
== Customize system instructions and parameters
== Shortcuts integration
== No subscriptions, buy once and get any update

Read the terms and privacy policy before using this app, available on latentlake.com/termspolicy/chat-policy-terms

Added support for the new Gemma 4.
Available in the model manager:
- Gemma4 E2B + vision
- Qwen3.5 4B + vision

Disabled automatic reasoning for some models that reason by default.