AMV-AIChain
Why LangChain on Termux? Bloat & Crash vs 94KB Zero-Dep LCEL & Hardware Native Agents
Install the official package directly into your runtime:
pip install amv-aichain
# or:
npm install amv-aichain
The Engineering Challenge
Standard enterprise frameworks (LangChain, LlamaIndex, CrewAI) pull in 45+ heavy dependencies (Pydantic-core, Chroma, Tiktoken) weighing 200MB+ that constantly fail to compile on Android Bionic ARM64 and trigger instant Out-Of-Memory (LMK) kernel kills.
The Architectural Breakthrough
AMV-AIChain provides 100% pure standard-library LCEL pipe chaining (prompt | llm | parser), autonomous StateGraph agents, on-device SQLite vector RAG, and first-class Android smartphone hardware actuation (battery, temperature, haptics, GPS) in a tiny 94KB footprint.
Key Capabilities & Built-in Hardening
Zero External Dependencies (94KB)
Written entirely in Python stdlib (urllib, sqlite3, json) and Node.js ESM. No C/Rust compiler required on Android Bionic ARM64.
LCEL Pipeline Compatibility
Full Runnable Protocol support with the pipe operator (prompt | llm | parser) and real-time streaming chunk emission.
Native Smartphone Hardware Actuation
Built-in first-class tools for Android kernel battery, surface temperature, GPS coordinates, haptic vibration, and system notifications.
On-Device SQLite Vector Store (RAG)
Embeddings storage and pure-math cosine similarity search in SQLite :memory: or local DB without ChromaDB or FAISS.
Supported Compute Kernels & Operations
| Subsystem Category | Operations & Kernels | Status |
|---|---|---|
| LCEL Pipe Protocol | PromptTemplate | Model | OutputParser Runnable Sequence | Production |
| Autonomous Agents | StateGraph, create_react_agent, ToolPolicy(default='deny') | Production |
| Hardware Actuation | Battery, Temperature, GPS, Vibration, Notifications, Shell | Production |
| Vector Store & RAG | SQLiteVectorStore with Pure Math Cosine Similarity | Production |
| Dual Runtime Parity | Pure Python 3.10+ Stdlib & Pure Node.js 18+ ESM | Production |
Canonical Usage Example
from amv_aichain import PromptTemplate, OpenAICompatibleChat, StringOutputParser, get_battery_status
# 1. Read physical smartphone hardware telemetry
battery = get_battery_status()
# 2. Build LCEL Chain connected to local llama-server (port 8080)
prompt = PromptTemplate.from_template("Phone Telemetry: {status}. Diagnose in 1 sentence:")
llm = OpenAICompatibleChat(base_url="http://127.0.0.1:8080/v1", model="llama3", temperature=0.3)
chain = prompt | llm | StringOutputParser()
# 3. Execute with 0 external dependencies
report = chain.invoke({"status": battery})
print("AI Diagnostic:", report)