Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/TrinaxCode/TrinaxAI/llms.txt

Use this file to discover all available pages before exploring further.

TrinaxAI is an open-source, local-first AI assistant that runs entirely on your own machine. It pairs a ChatGPT-like Progressive Web App with a developer CLI, semantic code search powered by Retrieval-Augmented Generation (RAG), voice conversations, and local image analysis — all without a cloud account, API key, or subscription. Your code, documents, and chat history never leave your network.

​
Architecture overview

TrinaxAI is a three-tier local stack where every service runs on your device: 
┌──────────────────────────────────────────┐
│              Your Device                 │
│  ┌──────────┐  ┌─────────────────────┐   │
│  │PWA(React)│  │  CLI (trinaxai)     │   │
│  │  :3334   │  │  pip install -e .   │   │
│  └─────┬────┘  └──────────┬──────────┘   │
│        │                  │              │
│  ┌─────┴──────────────────┴──────────┐   │
│  │    RAG API (FastAPI) :3333        │   │
│  │  LlamaIndex · bge-m3 · BM25      │   │
│  └─────┬─────────────────────────────┘   │
│        │                                 │
│  ┌─────┴──────┐                          │
│  │   Ollama   │  qwen2.5 · llama3.2     │
│  │   :11434   │  bge-m3 · moondream     │
│  └────────────┘                          │
└──────────────────────────────────────────┘
TierComponentPortRole
FrontendReact 19 PWA (TypeScript + Vite):3334Chat UI, voice, vision, PWA install
BackendFastAPI + LlamaIndex RAG API:3333Hybrid retrieval, memory, streaming
ModelsOllama model runtime:11434LLM inference, embeddings
Your data stays on your machine. TrinaxAI makes no outbound network requests to any cloud service. The only external call in the PWA is to load Google Fonts. No chat, code, or documents are ever uploaded anywhere.

​
Key features

RAG & Code Indexing

Index your projects for semantic search with citations. AST-aware chunking for 15+ languages, hybrid vector + BM25 retrieval, and incremental re-indexing that only touches changed files.

Chat & Models

Chat via the PWA or CLI with an auto-routing heuristic that picks the best Ollama model for each query. Deep research mode decomposes questions across multiple RAG passes.

Voice & Vision

Full voice conversations with speech-to-text and text-to-speech, plus local image and screenshot analysis powered by qwen2.5vl:3b — no third-party API required.

Memory & Collections

Persistent memory stores “remember that…” facts locally and syncs across devices. Knowledge Collections let you create separate RAG namespaces and query one or many at once.

Progressive Web App

Install TrinaxAI as a native app on iOS, Android, or desktop. Served over self-signed HTTPS for LAN access. Supports dark/light mode and Spanish/English auto-detection.

Security Model

LAN system control is disabled by default. RAG API, Ollama, and the PWA are all localhost-bound unless you explicitly enable LAN access. Full threat model and hardening guide included.

​
Supported platforms

OSInstallerService Manager
Linux (Ubuntu, Debian, Fedora, Arch)install.shuser systemd
macOS (Intel + Apple Silicon)install.shlaunchctl
Windows (10/11, PowerShell)install.ps1subprocess supervisor
Platform-specific guides: Linux · macOS · Windows

​
Hardware profiles

The installer auto-detects your available RAM and selects the best model profile. You can override it with --profile <name> during install or by setting TRINAXAI_PROFILE in .env.
ProfileRAM TargetGeneral modelCode modelDeep modelEmbedding
8gb~8 GBllama3.2:1bqwen2.5-coder:1.5bqwen2.5-coder:1.5bnomic-embed-text
16gb~16 GBllama3.2:3bqwen2.5-coder:3bqwen2.5-coder:3bbge-m3
max32 GB+llama3.2:3bqwen2.5-coder:3bqwen2.5-coder:7bbge-m3
ultra64 GB+llama3.2:3bqwen2.5-coder:3bqwen2.5-coder:14bbge-m3
The available embedding models are bge-m3 (balanced, multilingual), nomic-embed-text (lite, faster), and all-minilm (fast, smallest). The profile sets the default preset, which you can override with TRINAXAI_EMBED_PRESET in .env. The auto-router selects among MODEL_GENERAL, MODEL_CODE, MODEL_DEEP, and MODEL_FAST at query time using a heuristic classifier — no extra LLM call needed.

​
What’s included

  • Developer CLI — trinaxai ask, trinaxai chat, trinaxai index, trinaxai browse, trinaxai doctor, trinaxai research, trinaxai memory, trinaxai collections, trinaxai watch, trinaxai export, trinaxai obsidian
  • PWA — 18 TypeScript components, session history, search, export to Markdown/PDF/Word, in-app docs, 7-step onboarding wizard
  • Bilingual UI — Spanish and English, auto-detected from browser locale
  • One-command installers — install.sh (Linux/macOS) and install.ps1 (Windows), with guided update and uninstall scripts
  • Continue.dev integration — VSCode config included for IDE-native AI completions
Ready to install? Head to the Quickstart for a step-by-step walkthrough.

Build docs developers (and LLMs) love

Get started for freeTalk to us