Open Source Foundation

Intelligence at the
Edge of Everything

Andyria is a self-improving, locally-operated AI agent platform. No cloud dependency. No per-token bill. Just autonomous intelligence running where you need it — on your hardware, in your mesh.

⭐ Star on GitHub Explore the Platform
Local‑First
Zero cloud required
Self‑Learning
Grows smarter every session
Open
Apache 2.0 licensed
50% Donated
Profits → CFF research

An AI agent that builds itself

Every component of Andyria is designed to operate offline-first, improve from its own outputs, and extend itself without human intervention.

🧠

Chain-of-Thought Reasoning

ReasoningEngine decomposes complex prompts into sub-questions, analyzes each independently, then synthesizes a coherent answer — all locally.

📚

Auto-Learn Loop

High-confidence responses are automatically distilled into MEMORY.md. Future prompts inject this learned context, creating a compounding intelligence effect.

🔄

ATM — Automated Thought Machine

Iterative generate → critique → revise loop with configurable depth. Escalates to full chain-of-thought reasoning when confidence falls below threshold.

🌐

Mesh Networking

Peer-to-peer gossip protocol syncs DAG state across nodes. No central server. Runs on Raspberry Pi, cloud VMs, or your laptop simultaneously.

🔐

Cryptographic DAG Ledger

Every event is signed with Ed25519, hashed with BLAKE3, and appended to an immutable causal DAG. Full audit trail, zero trust required.

🎛️

Multi-Model Routing

Routes tasks to local GGUF models → Ollama → stub fallback. Zero cost path always attempted first. Bring your own weights.

Entropy-Anchored Identity

Physical entropy beacons anchor every request to real-world time and hardware state, making replay attacks cryptographically infeasible.

🤖

Multi-Agent Orchestration

Spawn, delegate, and chain specialized agents across a DAG execution graph. Each agent carries a persona, skill profile, and private memory.

🔋

Edge-Optimized Runtime

First-class Raspberry Pi support. Runs on 2GB RAM. Docker Compose deployment in under 60 seconds. Designed for Termux and ARM.

Built for autonomy,
not abstraction

A layered stack where every component has one job and emits signed events to a shared DAG. No hidden state. No black boxes.

🧩

Coordinator

Entropy → Plan → Route → Verify → Reflect → Learn

🔗

DAG Ledger

Causal graph of every signed event, Rust-native

🕸️

Mesh Layer

Gossip-based sync, no coordinator node

ANDYRIA RUNTIME v2.x
─────────────────────────────
Coordinator ← HTTP/WebSocket
├─ ReasoningEngine [CoT]
├─ AutomatedThoughtMachine
├─ AutoLearner → MEMORY.md
├─ ModelRouter [GGUF|Ollama]
├─ Planner + Verifier
└─ MeshManager [gossip]
EventDAG (Rust — BLAKE3 + Ed25519)
├─ store.rs / crypto.rs
└─ entropy.rs / event.rs
Memory Layer
├─ ContentAddressedMemory [BLAKE3]
├─ PersistentMemory [MEMORY.md]
└─ SessionStore [turns]
Nodes :7700 / :7701 ...

Up in 60 seconds

One command. No API keys, no cloud account, no credit card.

# One-liner installer (Linux / macOS / WSL / Termux / Raspberry Pi)
curl -fsSL https://andyriax.github.io/andyria/install.sh | bash

# With options: pick a preset agent, custom port, Python-only mode
curl -fsSL https://andyriax.github.io/andyria/install.sh | bash -s -- \
--agent coder --port 7700 --python

# Or clone and run manually
git clone https://github.com/andyriax/andyria.git && cd andyria
docker compose up -d # → http://localhost:7700

# Optional: connect Ollama for a free local LLM
ollama pull llama3
# Andyria auto-detects it — no config needed

Intelligence as a human right

"We believe that powerful AI should not require a cloud account, a subscription, or a data center. Andyria exists to prove that autonomous intelligence can run on the hardware you already own, improve itself without external intervention, and remain fully under your control."

The Andyria Foundation is committed to open-source development, radical transparency, and a governance model that puts community benefit above commercial interest. All core technology is and will remain freely available under the Apache 2.0 license.

💛

Supporting the Cystic Fibrosis Foundation

50% of all Andyria Foundation profits are donated directly to CFF research. Every deployment, every commercial license, every sponsored feature — half goes to fighting cystic fibrosis. This is a founding commitment, not a marketing line.

Where we're going

Built in the open. Every milestone is a GitHub release.

v1 — Core DAG + Mesh Shipped

Cryptographic event DAG, Ed25519 identity, gossip mesh, ModelRouter, ATM think/reflect loop.

v1.5 — Agent Platform Shipped

Multi-agent registry, persona archetypes, skill profiles, session memory, TodoStore, CronScheduler, delegation chains.

v2 — Reasoning + Auto-Learn Active

ReasoningEngine (CoT decompose→analyze→synthesize), AutoLearner (pattern distillation into MEMORY.md), dynamic Surprise Me, new event types.

v2.5 — TT Live Agent Next

TikTok Live monetization agent, JETS token rewards, OpenClaw protocol, revenue skill chain.

v3 — Distributed Swarm Planned

Multi-node swarm coordination, Rust runtime for ARM, WASM agent sandboxing, federated learning across mesh.

v4 — Foundation Platform Planned

Community governance, grant program, SDK for third-party agent authors, foundation membership model.

Build with us

Andyria is built in the open. Whether you contribute code, documentation, agents, or ideas — you're part of the foundation.

⭐ Star the Repo 🍴 Fork & Contribute 🐛 Open an Issue