What is NeuralCleave?
NeuralCleave is an open-source personal AI assistant runtime that connects a large language model to 32 messaging platforms, 13 AI providers, local voice, self-modifying skills, and a privacy mode that keeps every byte on your machine. Unlike tools that wrap an API behind a chatbot UI, NeuralCleave runs the model stack in-process — HuggingFace, faster-whisper, Kokoro TTS, OpenWakeWord, and sentence-transformers all execute locally alongside the Python gateway.Key Features
32 Messaging Channels
Telegram, Discord, Slack, WhatsApp, Email, SMS, and 26 more. One gateway, all platforms.
13 AI Providers
OpenAI, Anthropic, Gemini, DeepSeek, Ollama, Mistral, xAI Grok, Cohere, and more — switchable at runtime.
Local Voice Pipeline
Wake word (OpenWakeWord) → VAD → STT (faster-whisper) → LLM → TTS (ElevenLabs / Kokoro / pyttsx3). 100% local when using Ollama + Kokoro.
Self-Modifying Skills
Skills are Python files the agent can read, write, and reload at runtime. The assistant extends itself.
Semantic Memory
Vector-embedded conversation memory with cosine similarity retrieval. In-process, no external vector DB required.
Privacy Mode
One flag routes everything — LLM, STT, TTS, embeddings — through local models. Zero external API calls.
How it compares
Where to start
Quickstart
Get a live agent in under five minutes.
Configuration
Full
config.toml reference.Voice Setup
Enable wake word, STT, and TTS.
API Reference
REST API for all gateway operations.