restore original dervish.gif below explanation; add new logo left-aligned, smaller

README.md

@@ -1,6 +1,6 @@
 # Dervish MCP
 
-<p align="center"><img src="dervish.gif" alt="Dervish"></p>
+<p align="left"><img src="dervish-logo.png" alt="Dervish" width="120"></p>
 
 **Dervish** infers **regular expression grammars** from example sequences using the BEX family of algorithms. Given a set of example sequences (strings over some alphabet), it learns a compact regular expression that captures the general pattern.
 
@@ -15,6 +15,8 @@ Dervish replaces both with a **one-call MCP tool**: pass your sequences, get bac
 
 **Without Dervish:** token cost scales linearly with examples. **With Dervish:** one compact grammar describes them all — a ~60–200 token rule instead of thousands of tokens of raw examples.
 
+<p align="center"><img src="dervish.gif" alt="Dervish animation" width="75%"></p>
+
 ## MCP Server
 
 The primary interface is a **Model Context Protocol (MCP)** server. Connect any MCP-compatible client (pi.dev, opencode, vibe, etc.) and get grammar inference as a tool:

SHOWCASE.md

@@ -1,6 +1,6 @@
 # Dervish — Showcase
 
-<p align="center"><img src="dervish.gif" alt="Dervish" width="200"></p>
+<p align="left"><img src="dervish-logo.png" alt="Dervish" width="120"></p>
 
 Infer the **unwritten convention** from existing examples. Given N example
 sequences, produce a ~100-char grammar that captures the structural

blog_post.md

@@ -1,6 +1,6 @@
 # Dervish: Discovering Unwritten Conventions with Grammar Inference
 
-<p align="center"><img src="dervish.gif" alt="Dervish" width="200"></p>
+<p align="left"><img src="dervish-logo.png" alt="Dervish" width="120"></p>
 
 **How we turned 36 Ansible roles into a 200-character grammar — and why
 it matters for LLM agents.**