🪹 Bird's Nest

The First Local AI Manager for Non-Transformer Models

Run RWKV, Mamba, xLSTM, and StripedHyena locally on your Mac.
No cloud. No API keys. No transformers.

---

Why Bird's Nest?

Every local AI tool — Ollama, LM Studio, GPT4All — is built for transformers. We built for what comes next.

	Transformer Tools	Bird's Nest
Architecture	Attention-based (GPT, LLaMA, Mistral)	RNN, SSM, LSTM (RWKV, Mamba, xLSTM)
Memory	KV cache grows with context	Constant memory
Scaling	O(n²) attention	O(1) per token
Context	Fixed window (4K–128K)	Infinite context
RAM for 7B	16GB+ recommended	8GB capable

---

Screenshots

Chat with RWKV models — streaming responses at 11.5 tok/s

25+ tools your local AI can call — web search, image gen, music, YouTube, and more

One-click model management — download, load, and switch between architectures

On-device image generation — your RNN calls tools mid-conversation to create images locally

---

Features

🧠 Multi-Architecture AI Engine

Run 19 text models across 4 non-transformer architectures — all from one app:

Family	Models	Parameters	Architecture
RWKV-7 GooseOne	7 variants	0.1B – 13.3B	Linear Attention RNN
RWKV-7 World	4 variants	0.1B – 2.9B	Multilingual RNN
RWKV-6 Finch	4 variants	1.6B – 14B	Previous Gen RNN
Mamba	2 variants	1.4B – 2.8B	Selective State Space
xLSTM	1 variant	7B	Extended LSTM
StripedHyena	1 variant	7B	Hybrid Architecture

🎨 Image Generation

8 image models with per-model quantization (Q4/Q8) and TeaCache acceleration:

• SDXL Lightning — 4-step instant generation
• FLUX.2 Klein — 4B and 9B variants
• FLUX.1 Dev — 12B with LoRA and ControlNet
• Qwen Image 20B — best prompt understanding
• Z-Image Turbo — photorealism specialist
• FLUX.1 Kontext — style transfer and editing
• SeedVR2 — 1-step upscaler

🔧 25+ Built-in Tools

Your local RNN can call tools mid-conversation — no cloud APIs needed:

Tool	Description
🔍 Web Search	DuckDuckGo search with inline results
🖼️ Image Search	Find and display images
📺 YouTube	Search and embed videos inline
🌐 Fetch URL	Read and summarize web pages
🎨 Image Gen	Generate images from text prompts
🎵 Music Gen	Create music with Stable Audio / Riffusion
🔄 Translate	50+ language pairs
📚 RAG	Search local documents
🔢 Calculator	Math and conversions
🌤️ Weather	Current conditions
📝 Todos	Task management
🐍 Python	Run code inline
📁 File Search	Search local files
🖥️ System	System monitoring
...and more	Screenshot, Clipboard, Shell, Database, Upscale, Edit Image

🎵 Music Generation

• Stable Audio Open — up to 47s of high-quality audio
• Riffusion — real-time music from text prompts

🔒 100% Local

• Zero cloud dependencies — everything runs on your machine
• Zero telemetry — no analytics, no tracking, no phone-home
• Zero API keys — download models from HuggingFace and go
• Your data stays yours — chat history, images, everything is local

---

Quick Start

# Clone the repo
git clone https://github.com/Dappit-io/birdsnest.git
cd birdsnest

# Create virtual environment
python3 -m venv .venv
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Launch Bird's Nest
python -m birdsnest.server

Open http://localhost:7861 in your browser. That's it.

First Run

1. Click any model in the Model Library panel
2. Click Download — models auto-download from HuggingFace
3. Once downloaded, click the model to Load it
4. Start chatting!

---

System Requirements

Requirement	Minimum	Recommended
OS	macOS 12+	macOS 14+
CPU	Any (Intel/Apple Silicon)	Apple Silicon (M1+)
RAM	8 GB	32 GB
Python	3.11	3.12+
Disk	5 GB (smallest model)	50 GB (full catalog)

💡 Apple Silicon users get Metal acceleration for image generation via MLX. Intel Macs work fine for text models.

---

Architecture

Browser (localhost:7861)
    │
    ├── WebSocket /ws/chat ──► Streaming token generation
    │
    └── REST API
         ├── /api/models ──────► Model download, load, switch
         ├── /api/image-models ─► Image model management
         ├── /api/image-library ► Generated image gallery
         ├── /api/music-models ─► Music model management
         └── /api/tools ────────► Tool execution pipeline

Server (FastAPI + Uvicorn)
    │
    ├── engines/
    │    ├── rwkv_engine.py ───► RWKV v6 & v7 inference
    │    ├── mamba_engine.py ──► Mamba SSM inference
    │    └── hf_engine.py ────► xLSTM, StripedHyena (HuggingFace)
    │
    ├── image_engine.py ──────► mflux + diffusers (8 models)
    ├── tools.py ─────────────► 25+ built-in tool implementations
    ├── rag.py ───────────────► Local RAG with sentence embeddings
    └── models.py ────────────► Model catalog definitions

---

Data Storage

Data	Location
Models	~/birdsnest_models/
Generated images	~/birdsnest_workspace/images/
Chat history	~/birdsnest_workspace/.birdsnest_history.json
Config	~/birdsnest_workspace/.birdsnest_*

---

Contributing

Pull requests welcome! See CONTRIBUTING.md for guidelines, development setup, and how to add new models and tools.

Changelog

See CHANGELOG.md for release history.

License

MIT — use it, fork it, build on it.

---

Built by Dappit — The First Web3 Vibe Coding Platform
_{Build, deploy, and ship decentralized apps with AI — no setup required.}

dappit.io · GitHub