Update README.md

README.md

@@ -1,10 +1,104 @@
----
-title: README
-emoji: 📉
-colorFrom: purple
-colorTo: purple
-sdk: static
-pinned: false
+![Minsore logo](logo.png)
+
+# Minsore — Small models. Serious intelligence.
+
+**Minsore** is a company building **small, efficient coding models** that are easy to run **locally** — without massive GPUs or expensive infrastructure.
+
+We focus on compact models tuned for real developer work: code generation, debugging, agents, and software tasks. Less memory. Cheaper inference. Deploy anywhere — from laptop to cloud API.
+
+One family. Three scales:
+
+`0.7B → 1.7B → 3B`
+
+## Why Minsore
+
+Big models aren't the only way forward.
+
+- **Less memory.** Run where large systems can't fit.
+- **Cheaper inference.** Lower cost per task, by design.
+- **Accessible hardware.** No exotic cluster required.
+- **Local + API.** From laptop to production endpoint.
+- **Developer-first.** Built around coding, debugging and software tasks.
+
+> Intelligence doesn't have to be enormous. Useful intelligence should fit where people need it.
+
+## Wing family
+
+A family of compact coding models built to scale with the task.
+
+| Model | Size | Best for |
+|---|---|---|
+| **Wing Nano** `wing-nano` | ~0.7B | Fast, inexpensive coding assistance, lightweight agents, environments where every GB matters. Low memory, fast inference, ultra-low cost. |
+| **Wing Junior** `wing-junior` | ~1.7B | Everyday coding model for deeper reasoning, debugging, code generation and tool use. Balance of capability and cost. |
+| **Wing Senior** `wing-senior` | ~3B | Most capable Wing model. Advanced coding, complex reasoning, agentic workflows. Compact flagship pushing toward much larger systems quality. |
+
+Capability grows with scale — no benchmark claims without measured results.
+
+### Choose your weight class
+
+- **Wing Nano ~0.7B** — low cost
+- **Wing Junior ~1.7B** — balance
+- **Wing Senior ~3B** — maximum capability
+
+More capability when you need it. Less hardware when you don't.
+
+Start small. Scale only when the task demands it.
+
+## Local-first
+
+All Wing models are designed to be **local-friendly**:
+
+- Run on consumer hardware
+- Quantizable for low-memory setups
+- Ideal for Ollama / llama.cpp / Transformers / vLLM style workflows
+- Prototype locally, deploy cheaply
+
+## API — Built for cheap inference
+
+Use Wing through a simple **OpenAI-compatible API** — swap `base_url`, keep your client. No GPUs to manage.
+
+Planned starting price for `wing-nano`:
+- **$0.02 / 1M input tokens**
+- **$0.09 / 1M output tokens**
+
+Start on `wing-nano`, upgrade to Junior / Senior without rewrites.
+
+```python
+from openai import OpenAI
+
+client = OpenAI(
+    base_url="https://api.minsore.com/v1",
+    api_key="YOUR_API_KEY"
+)
+
+response = client.chat.completions.create(
+    model="wing-nano",
+    messages=[
+        {"role": "user", "content": "Write a Python binary search function."}
+    ]
+)
+
+print(response.choices[0].message.content)
+```
+
+## Developer workflow: Prompt → Wing → Code → Test → Fix
+
+A tight loop, not a black box:
+
+1. **Generate** — Wing generates code from your prompt and context.
+2. **Verify** — Code runs in a verifier — tests execute, errors are captured.
+3. **Correct** — Failures return to the model. Wing revises the solution.
+
+## Roadmap
+
+Staged, not rushed.
+
+1. **Wing Nano** — In progress — Compact coding intelligence.
+2. **Wing Junior** — A stronger everyday developer model.
+3. **Wing Senior** — The compact flagship.
+4. **Verified Runtime** — Prototype — Generation, testing and automatic correction.
+5. **Minsore API** — Simple, affordable inference for developers.
+
 ---
 
-Edit this `README.md` markdown file to author your organization card.
+© 2026 Minsore — Minsore → Wing → Nano / Junior / Senior