diff --git a/src/changeloggen/changelog_generator.py b/src/changeloggen/changelog_generator.py
new file mode 100644
index 0000000..4f2451f
--- /dev/null
+++ b/src/changeloggen/changelog_generator.py
@@ -0,0 +1,253 @@
+from datetime import datetime
+from typing import Optional
+from rich.console import Console
+from rich.panel import Panel
+from rich.text import Text
+from rich.table import Table
+from rich.markdown import Markdown
+
+from .git_client import ChangeSet, FileChange
+from .llm_client import CategorizedChange, LLMResponse, OllamaAPIClient, build_categorization_prompt, Config as LLMConfig
+
+
+CONVENTIONAL_TYPES = {
+    'feat': ('Features', '✨'),
+    'fix': ('Bug Fixes', '🐛'),
+    'docs': ('Documentation', '📝'),
+    'breaking': ('Breaking Changes', '💥'),
+    'refactor': ('Refactoring', '♻️'),
+    'style': ('Styles', '💄'),
+    'test': ('Tests', '✅'),
+    'chore': ('Chore', '🔧'),
+}
+
+
+def categorize_changes(
+    changes: ChangeSet,
+    llm_client: Optional[OllamaAPIClient] = None,
+    model: str = "llama3.2"
+) -> LLMResponse:
+    """Categorize git changes using LLM."""
+    if llm_client is None:
+        config = LLMConfig(model=model)
+        llm_client = OllamaAPIClient(config)
+    
+    changes_text = _format_changes_for_llm(changes)
+    prompt = build_categorization_prompt(changes_text)
+    
+    response_text = llm_client.generate(prompt)
+    
+    return _parse_llm_response(response_text)
+
+
+def _format_changes_for_llm(changes: ChangeSet) -> str:
+    """Format changes into text for LLM processing."""
+    lines = []
+    
+    if changes.staged_changes:
+        lines.append("=== STAGED CHANGES ===")
+        for change in changes.staged_changes:
+            lines.append(f"\nFile: {change.file_path}")
+            lines.append(f"Type: {change.change_type}")
+            lines.append(f"Changes:\n{change.diff_content}")
+    
+    if changes.unstaged_changes:
+        lines.append("\n=== UNSTAGED CHANGES ===")
+        for change in changes.unstaged_changes:
+            lines.append(f"\nFile: {change.file_path}")
+            lines.append(f"Type: {change.change_type}")
+            lines.append(f"Changes:\n{change.diff_content}")
+    
+    return '\n'.join(lines)
+
+
+def _parse_llm_response(response_text: str) -> LLMResponse:
+    """Parse LLM response into structured data."""
+    import json
+    
+    try:
+        json_start = response_text.find('{')
+        json_end = response_text.rfind('}') + 1
+        json_str = response_text[json_start:json_end]
+        data = json.loads(json_str)
+        
+        changes = []
+        for item in data.get('changes', []):
+            changes.append(CategorizedChange(
+                type=item.get('type', 'chore'),
+                scope=item.get('scope'),
+                description=item.get('description', ''),
+                file_path=item.get('file_path', ''),
+                breaking_change=item.get('breaking_change', False),
+                breaking_description=item.get('breaking_description'),
+                confidence=item.get('confidence', 0.0)
+            ))
+        
+        return LLMResponse(
+            changes=changes,
+            summary=data.get('summary', ''),
+            version=data.get('version', '1.0.0'),
+            breaking_changes=data.get('breaking_changes', []),
+            contributors=data.get('contributors', [])
+        )
+    except (json.JSONDecodeError, AttributeError, KeyError) as e:
+        return LLMResponse(
+            changes=[],
+            summary=f"Failed to parse LLM response: {str(e)}",
+            breaking_changes=[]
+        )
+
+
+def group_changes_by_type(changes: list[CategorizedChange]) -> dict[str, list[CategorizedChange]]:
+    """Group categorized changes by type."""
+    grouped = {}
+    for change in changes:
+        change_type = change.type.lower()
+        if change_type not in grouped:
+            grouped[change_type] = []
+        grouped[change_type].append(change)
+    return grouped
+
+
+def format_conventional_changelog(
+    categorized: LLMResponse,
+    version: str = "1.0.0",
+    date: Optional[str] = None
+) -> str:
+    """Generate markdown changelog following conventional commits format."""
+    if date is None:
+        date = datetime.now().strftime("%Y-%m-%d")
+    
+    lines = []
+    lines.append(f"# Changelog v{version} ({date})")
+    lines.append("")
+    
+    if categorized.breaking_changes:
+        lines.append("## 💥 Breaking Changes")
+        lines.append("")
+        for bc in categorized.breaking_changes:
+            lines.append(f"- {bc}")
+        lines.append("")
+    
+    grouped = group_changes_by_type(categorized.changes)
+    
+    for change_type, (display_name, emoji) in CONVENTIONAL_TYPES.items():
+        if change_type == 'breaking':
+            continue
+        if change_type in grouped:
+            lines.append(f"## {emoji} {display_name}")
+            lines.append("")
+            for change in grouped[change_type]:
+                scope_part = f"({change.scope})" if change.scope else ""
+                lines.append(f"- **{change.type}{scope_part}:** {change.description}")
+            lines.append("")
+    
+    if categorized.summary:
+        lines.append("---")
+        lines.append("")
+        lines.append("### Summary")
+        lines.append("")
+        lines.append(categorized.summary)
+    
+    return '\n'.join(lines)
+
+
+def format_json_output(categorized: LLMResponse, version: str = "1.0.0") -> str:
+    """Export changelog as structured JSON."""
+    import json
+    
+    output = {
+        "version": version,
+        "generated_at": datetime.now().isoformat(),
+        "summary": categorized.summary,
+        "changes": [
+            {
+                "type": c.type,
+                "scope": c.scope,
+                "description": c.description,
+                "file_path": c.file_path,
+                "breaking_change": c.breaking_change,
+                "breaking_description": c.breaking_description,
+                "confidence": c.confidence
+            }
+            for c in categorized.changes
+        ],
+        "breaking_changes": categorized.breaking_changes,
+        "contributors": categorized.contributors
+    }
+    
+    return json.dumps(output, indent=2)
+
+
+def format_release_notes(
+    categorized: LLMResponse,
+    version: str = "1.0.0",
+    repo_url: Optional[str] = None
+) -> str:
+    """Generate release notes for GitHub/GitLab."""
+    lines = []
+    lines.append(f"## Release v{version}")
+    lines.append("")
+    
+    if categorized.breaking_changes:
+        lines.append("### ⚠️ Breaking Changes")
+        lines.append("")
+        for bc in categorized.breaking_changes:
+            lines.append(f"- {bc}")
+        lines.append("")
+    
+    grouped = group_changes_by_type(categorized.changes)
+    
+    for change_type, (display_name, emoji) in CONVENTIONAL_TYPES.items():
+        if change_type == 'breaking':
+            continue
+        if change_type in grouped:
+            lines.append(f"### {emoji} {display_name}")
+            lines.append("")
+            for change in grouped[change_type]:
+                scope_part = f"**({change.scope})** " if change.scope else ""
+                lines.append(f"- {scope_part}{change.description}")
+            lines.append("")
+    
+    if categorized.contributors:
+        lines.append("### Contributors")
+        lines.append("")
+        lines.append("Thank you to our contributors:")
+        lines.append("")
+        for contributor in categorized.contributors:
+            lines.append(f"- {contributor}")
+        lines.append("")
+    
+    if categorized.summary:
+        lines.append("---")
+        lines.append("")
+        lines.append(f"**{categorized.summary}**")
+    
+    return '\n'.join(lines)
+
+
+def print_changelog_pretty(categorized: LLMResponse, console: Optional[Console] = None):
+    """Print changelog using Rich for pretty formatting."""
+    if console is None:
+        console = Console()
+    
+    grouped = group_changes_by_type(categorized.changes)
+    
+    for change_type, (display_name, emoji) in CONVENTIONAL_TYPES.items():
+        if change_type == 'breaking':
+            continue
+        if change_type in grouped:
+            table = Table(title=f"{emoji} {display_name}", show_header=True)
+            table.add_column("Scope", style="cyan")
+            table.add_column("Description", style="green")
+            table.add_column("File", style="yellow")
+            
+            for change in grouped[change_type]:
+                table.add_row(
+                    change.scope or "-",
+                    change.description,
+                    change.file_path
+                )
+            
+            console.print(table)
+            console.print()