StewMind 1.7B is live.
A tiny agentic AI built in Nigeria 🇳🇬 that runs on YOUR machine.
Big models cook for everyone. This one cooks for you.
Most AI today is a vending machine: you pay per token, you rent someone else's brain, and when the API bills spike, your product dies. We built the opposite. A small, sharp, agentic brain you download once and run forever.
What it does
→ Agent routing: understands intent and picks the right skill
→ Tool-calling: structured function calls, not vibes
→ ReAct reasoning: thinks, acts, observes, repeats
→ Memory ops: built for agent systems that remember
→ English + Nigerian Pidgin: it understands how we actually talk
→ GGUF ready: runs on your laptop, your mini-PC, even your phone
→ $0 API bills: no subscriptions, no rate limits, no surprise invoices
Run it in 10 lines
from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("Erog0291/stewmind-1.7b")
model = AutoModelForCausalLM.from_pretrained("Erog0291/stewmind-1.7b", torch_dtype="auto")
messages = [{"role": "user", "content": "Plan a workflow: fetch weather, then notify the user"}]
ids = tok.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True)
out = model.generate(ids, max_new_tokens=256, do_sample=False)
print(tok.decode(out[0][ids.shape[1]:], skip_special_tokens=True))
No GPU? No problem. The Q4_K_M GGUF is right in the repo — run it with llama.cpp, LM Studio, or Ollama. It sips RAM like jollof disappears at a party.
Why 1.7B is a feature, not a bug
Every agent developer I know has the same problem: the big models are brilliant but slow, expensive, and they burn your margin one tool call at a time.
A small model fine-tuned specifically for agentic work does three things better:
It routes in milliseconds — no 900ms round trip just to decide which tool to call
It stays focused — trained on agentic workflows, not on everything under the sun
It runs where your users are — offline, on-device, private data never leaves the machine
The future of agents isn't one giant brain in a datacenter. It's thousands of small brains living in your apps, your scripts, your hardware.
Built in Nigeria, for the world
StewMind comes out of the S.T.E.W project — we're building Nigerian AI infrastructure: speech models that hear Igbo, Yoruba, Hausa and Pidgin, agentic models that think locally, and tools that work where connectivity is expensive and unreliable.
If you're building agents in emerging markets, you know the pain: latency kills UX and API costs kill startups. Local-first AI is not a compromise here. It's the strategy.
Download it
🥣 Model + GGUF: https://huggingface.co/Erog0291/stewmind-1.7b
Remix it, fine-tune it, ship it in your product. If you build something with StewMind, drop a comment — I want to see what you cook. 🍲
And if you work with African languages, we also built Stew Yanga — Nigerian speech-to-text that actually understands Nigerian accents: https://huggingface.co/Erog0291/stew-yanga
Top comments (0)