Back to Meechi
Technical Specification

Tech Architecture

A mapping of the independent, local-first technical stack.

The core philosophy: Meechi treats your device as the absolute source of truth. Rather than pushing your data to a centralized cloud, we bring the intelligence down to your machine. The architecture is a modular ecosystem where the interface, the storage layer, and the AI models run entirely on your own hardware.

Absolute Ownership

The Meechi engine is built for speed, privacy, and absolute data ownership. No central cloud, no telemetry, no compromise. Every byte stays on your hardware.

Ecosystem

The complete distribution layer. Local-first logic available across all platforms without central silos.

Next.js PWA

A zero-install, browser-native logic layer built on React. Leveraging Dexie.js for IndexedDB persistence and service workers for background sync orchestration.

Tauri Desktop

The high-performance Rust wrapper that acts as the system glue. It orchestrates native disk synchronization, hardware acceleration, and manages local server sidecars.

Capacitor Mobile

Native mobile abstraction wrapping the web engine. Leverages your phone's native NPU for free, energy-efficient voice conversations (TTS/STT), and syncs securely via P2P LAN.

Intelligence

Multi-layer inference engine designed to provide high-fidelity intelligence while keeping every byte on your hardware.

llama.cpp

Native hardware inference. Provides maximum performance acceleration for offline desktop and mobile platforms.

WebLLM

Browser-native inference. Brings WASM and WebGPU-powered models directly to your browser for zero-install availability.

LlamaIndex RAG

Advanced RAG orchestration. Powered by LlamaIndex for high-fidelity knowledge retrieval, chunking, and distributed context management.

LLMLingua-2

Semantic prompt compression. High-density context optimization that allows for longer memories and more precise agentic reasoning.

Silero VAD

Enterprise-grade Voice Activity Detection running entirely in-browser via WebAssembly. Ensures precise microphone triggers without cloud latency.

Kokoro TTS

High-fidelity text-to-speech powered natively by ONNX Runtime Web. Achieves pristine, human-like voice synthesis entirely in-browser.

Whisper STT

Robust automatic speech recognition running isolated on your hardware. Transcribes speech-to-text without ever sending raw audio to the cloud.

Transformers.js

In-browser machine learning engine. Powers local embedding generation (all-MiniLM) for vector memory retrieval entirely offline.

Connectivity

Privacy-preserving communication protocols for data portability and real-time multimodal interaction.

MCP Protocol

Full Model Context Protocol integration. Connect your system, files, and external tools to your local agent with zero configuration.

WebRTC

Real-time sync for voice and video communication between devices.

Yjs CRDTs

Peer-to-peer data synchronization. Conflict-free replicated data types ensuring your offline edits merge flawlessly when you reconnect.

Syncthing

Encrypted, folder-to-folder synchronization.

Toolbelt

Studio-grade tools built on human-readable foundations for total documentation interoperability.

Markdown-First

Human-readable, portable, and standard-native storage. Your notes and thoughts belong to you, today and fifty years from now.

Semantic Data Injection

Our proudest storage design. We infuse rich metadata—like highlights, colors, and personal annotations—directly into your text using standard Markdown links. Your documents carry their own context without relying on fragile external databases.

LiteParse

Built-in lightweight document parsing for lightning-fast, on-the-fly text extraction.

Docling

Heavy-duty layout-aware structure extraction that turns complex PDFs and spreadsheets into highly structured local intelligence.

Tiptap

A studio-grade, local-first rich-text engine for the protagonist journey. Collaborative writing features without cloud dependency.

Agentic Records

Tiered memory systems designed for different cognitive tasks, from long-form writing to structural recall.

Scrivener

The orchestrator for narrative architecture and complex document structure. Maintaining structural integrity across vast writing projects.

Ghostwriter

The stylistic mirror. Specializing in creative drafting and mimicry to capture and replicate your specific authorial voice.

Librarian Memory

Specialized structural RAG agent. Manages the long-term context of your journey with deep semantic links across your archives.

The Wise Peer

Dialectical mentor agent designed for philosophical grounding, creative feedback, and critical thought partnership.

Core Strategy

The foundational philosophy that separates the Meechi engine from standard AI cloud platforms.

Biographical Memory Core

The BMC strategy: A 5-level narrative hierarchy that evolves your data from a mere 'Moment' into your lifelong 'Mythos'.

BYO Keys & Local-First

Zero-telemetry by default. Use your own hardware, your own keys, and your own storage. Intelligence that you fully own and control.

Human-Readable Storage

Your data is stored in human-readable formats, ensuring long-term accessibility and portability.

Privacy-First

No tracking. No analytics. No data collection. Your usage patterns and personal data stay with you.

Zero-Leash Data

100% data locality with zero telemetry. Your intelligence, unchained.

FounderTechPrivacy (Zero Data)TermsSupportLatest Updates
v1.2.2 — Independent Core
© 2026 Meechi. Studio of Two.