ne-agent 0.2.3

First open-source AI agent for Northeast Indian languages

NE-Agent

Open-source AI agent for Northeast Indian languages. Built on the NE-Stack by MWire Labs, Shillong.

---

Quickstart

pip install ne-agent

Install Ollama and pull a model:

ollama pull qwen2.5:1.5b
ollama serve

Then run:

ne-agent

---

What is NE-Agent?

NE-Agent is a terminal-based agentic assistant that speaks Northeast India's indigenous languages. It automatically detects the input language, then routes each query to the right tool — retrieval-augmented search, Khasi-English translation, or speech transcription — and generates responses, all running locally with no API keys required.

---

Powered by NE-Stack

Component	Role	Model
NE-LID	Language identification	fastText, 11 languages, 99.09% accuracy
Tool Router	Dynamic tool selection	qwen2.5:1.5b (search / translate / transcribe)
NE-Embed	Multilingual embeddings	LaBSE fine-tuned, 768-dim
FAISS	Vector retrieval	IndexFlatIP, cosine similarity
NLLB (Khasi)	Khasi-English translation	Fine-tuned NLLB
NE-ASR	Speech transcription	Whisper-medium, 8 languages
Ollama	Local LLM	qwen2.5:1.5b (default)

---

Supported Languages

Assamese · Khasi · Garo · Mizo · Meitei · Bodo · Kokborok · Nyishi · Nagamese · English

Translation currently supports Khasi. Transcription supports Khasi, Garo, Mizo, Nagamese, Kokborok, Assamese, Chakma, and Wancho.

---

Bring Your Own Corpus

The default corpus ships 500 monolingual sentences each for Assamese, Khasi, Mizo, and Garo. To use your own documents, replace the txt files in ne_agent/data/ with your own — one sentence per line.

---

Requirements

• Python 3.9+
• Ollama with qwen2.5:1.5b pulled
• GPU recommended for translation/transcription tools; search works on CPU

---

Links

• PyPI: pypi.org/project/ne-agent
• NE-Embed: huggingface.co/MWirelabs/ne-embed
• NE-LID: huggingface.co/MWirelabs/ne-lid
• NE-ASR: huggingface.co/MWirelabs/ne-asr
• MWire Labs: mwirelabs.com

---

License

CC-BY-4.0 — MWire Labs, Shillong, Meghalaya.