diff --git a/parsers/__init__.py b/parsers/__init__.py
index e69de29..2a10239 100644
--- a/parsers/__init__.py
+++ b/parsers/__init__.py
@@ -0,0 +1,6 @@
+from .base import ParserManager
+from .html_parser import HTMLDocParser
+from .pdf_parser import PDFDocParser
+from .email_parser import EmailDocParser
+from .cable_parser import CableDocParser
+from .text_parser import TextDocParser
diff --git a/parsers/base.py b/parsers/base.py
new file mode 100644
index 0000000..5df479a
--- /dev/null
+++ b/parsers/base.py
@@ -0,0 +1,161 @@
+"""Base parser with subprocess isolation for Mosaic document processing."""
+import os
+import logging
+import multiprocessing
+from multiprocessing import Process, Queue
+from typing import Optional
+from dataclasses import dataclass, field
+
+logger = logging.getLogger(__name__)
+
+# Default limits
+DEFAULT_MEMORY_LIMIT_MB = 512
+DEFAULT_TIMEOUT_SECONDS = 60
+
+
+@dataclass
+class ParseResult:
+    """Result from a parser subprocess."""
+    success: bool = False
+    text: str = ""
+    metadata: dict = field(default_factory=dict)
+    doc_type: str = ""
+    title: str = ""
+    char_count: int = 0
+    error: Optional[str] = None
+    needs_ocr: bool = False
+
+
+class ParserManager:
+    """Manages document parsing with subprocess isolation.
+
+    All parsers run in subprocess workers with:
+    - Memory limits (resource.setrlimit)
+    - Timeouts
+    - No access to config, SQLite, or network
+    """
+
+    def __init__(self, memory_limit_mb: int = DEFAULT_MEMORY_LIMIT_MB,
+                 timeout_seconds: int = DEFAULT_TIMEOUT_SECONDS,
+                 min_chars_per_page: int = 50,
+                 max_pdf_pages: int = 200):
+        self.memory_limit_mb = memory_limit_mb
+        self.timeout_seconds = timeout_seconds
+        self.min_chars_per_page = min_chars_per_page
+        self.max_pdf_pages = max_pdf_pages
+        self._parsers = {}
+        self._register_parsers()
+
+    def _register_parsers(self):
+        """Register available parsers by doc_type."""
+        from .html_parser import HTMLDocParser
+        from .pdf_parser import PDFDocParser
+        from .email_parser import EmailDocParser
+        from .cable_parser import CableDocParser
+        from .text_parser import TextDocParser
+
+        self._parsers = {
+            'html': HTMLDocParser(),
+            'pdf': PDFDocParser(max_pages=self.max_pdf_pages,
+                               min_chars_per_page=self.min_chars_per_page),
+            'email': EmailDocParser(),
+            'cable': CableDocParser(),
+            'text': TextDocParser(),
+        }
+
+    def parse(self, file_path: str, doc_type: str,
+              content_hash: Optional[str] = None) -> ParseResult:
+        """Parse a document in a subprocess worker.
+
+        Args:
+            file_path: Path to the document file
+            doc_type: Document type (html, pdf, email, cable, text)
+            content_hash: Expected hash for integrity verification
+
+        Returns:
+            ParseResult with extracted text and metadata
+        """
+        if not os.path.exists(file_path):
+            return ParseResult(error=f"File not found: {file_path}")
+
+        # Verify hash before parsing
+        if content_hash:
+            from utils.cache import DownloadCache
+            actual_hash = DownloadCache.compute_hash(file_path)
+            if actual_hash != content_hash:
+                return ParseResult(
+                    error=f"Hash mismatch: expected {content_hash}, got {actual_hash}"
+                )
+
+        parser = self._parsers.get(doc_type)
+        if not parser:
+            # Try to detect type
+            parser = self._detect_parser(file_path)
+            if not parser:
+                return ParseResult(error=f"No parser for doc_type: {doc_type}")
+
+        # Run in subprocess
+        result_queue = Queue()
+        proc = Process(
+            target=self._worker,
+            args=(parser, file_path, result_queue, self.memory_limit_mb)
+        )
+        proc.start()
+        proc.join(timeout=self.timeout_seconds)
+
+        if proc.is_alive():
+            proc.terminate()
+            proc.join(timeout=5)
+            if proc.is_alive():
+                proc.kill()
+            return ParseResult(error=f"Parser timed out after {self.timeout_seconds}s")
+
+        if result_queue.empty():
+            return ParseResult(error="Parser process returned no result")
+
+        result = result_queue.get_nowait()
+
+        # Check for OCR needed (low char count in PDFs)
+        if doc_type == 'pdf' and result.success:
+            if result.char_count > 0 and result.metadata.get('page_count', 1) > 0:
+                chars_per_page = result.char_count / result.metadata.get('page_count', 1)
+                if chars_per_page < self.min_chars_per_page:
+                    result.needs_ocr = True
+                    logger.info("Document may need OCR: %.1f chars/page", chars_per_page)
+
+        return result
+
+    @staticmethod
+    def _worker(parser, file_path: str, result_queue: Queue, memory_limit_mb: int):
+        """Subprocess worker that runs the parser with resource limits."""
+        try:
+            # Set memory limit
+            try:
+                import resource
+                mem_bytes = memory_limit_mb * 1024 * 1024
+                resource.setrlimit(resource.RLIMIT_AS, (mem_bytes, mem_bytes))
+            except (ImportError, ValueError, OSError) as e:
+                # resource module may not be available on all platforms
+                pass
+
+            result = parser.parse(file_path)
+            result_queue.put(result)
+        except MemoryError:
+            result_queue.put(ParseResult(error="Parser exceeded memory limit"))
+        except Exception as e:
+            result_queue.put(ParseResult(error=f"Parser error: {str(e)}"))
+
+    def _detect_parser(self, file_path: str) -> Optional[object]:
+        """Try to detect the right parser from file content."""
+        try:
+            from utils.sanitize import validate_content_type
+            content_type = validate_content_type(file_path)
+            if 'html' in content_type:
+                return self._parsers['html']
+            elif 'pdf' in content_type:
+                return self._parsers['pdf']
+            elif 'text' in content_type:
+                return self._parsers['text']
+        except Exception:
+            pass
+        return self._parsers.get('text')
diff --git a/parsers/cable_parser.py b/parsers/cable_parser.py
new file mode 100644
index 0000000..e7f8904
--- /dev/null
+++ b/parsers/cable_parser.py
@@ -0,0 +1,228 @@
+"""Diplomatic cable parser for Mosaic — structured field extraction.
+
+Cables have a specific format with fields like SUBJECT, ORIGIN,
+CLASSIFICATION, TAGS, and a multi-section body.
+"""
+import re
+import logging
+from typing import Optional
+
+from .base import ParseResult
+
+logger = logging.getLogger(__name__)
+
+# Cable field patterns
+CABLE_FIELDS = {
+    'VZCZCXYZ': 'transmission_id',
+    'DE ': 'from_code',
+    'ZNR': 'precedence',
+    'ZNY': 'precedence',
+    'FM ': 'from',
+    'TO ': 'to',
+    'INFO ': 'info',
+    'BT': 'break',
+    'UNCLAS': 'classification',
+    'CONFIDENTIAL': 'classification',
+    'SECRET': 'classification',
+    'SIPDIS': 'distribution',
+    'E.O.': 'executive_order',
+    'TAGS:': 'tags',
+    'SUBJECT:': 'subject',
+    'REF:': 'reference',
+    'REFS:': 'reference',
+}
+
+# Classification levels
+CLASSIFICATIONS = ['UNCLASSIFIED', 'CONFIDENTIAL', 'SECRET', 'SECRET//NOFORN',
+                   'CONFIDENTIAL//NOFORN', 'UNCLAS', 'UNCLAS SECTION']
+
+# Section markers in cable bodies
+SECTION_PATTERN = re.compile(r'^\d+\.\s+\((?:U|C|S|SBU)\)', re.MULTILINE)
+
+# ANSI escape pattern
+ANSI_ESCAPE = re.compile(r'\x1b\[[0-9;]*[a-zA-Z]|\x1b\].*?\x07')
+
+
+class CableDocParser:
+    """Parse diplomatic cables into structured fields."""
+
+    def parse(self, file_path: str) -> ParseResult:
+        """Parse a diplomatic cable.
+
+        Extracts structured fields (SUBJECT, ORIGIN, CLASSIFICATION, TAGS, etc.)
+        as discrete fields BEFORE any chunking.
+        """
+        try:
+            with open(file_path, 'r', encoding='utf-8', errors='replace') as f:
+                content = f.read()
+
+            # Check if this is HTML (WikiLeaks format) or raw cable
+            if '<html' in content.lower()[:200] or '<!doctype' in content.lower()[:200]:
+                return self._parse_html_cable(content)
+            else:
+                return self._parse_raw_cable(content)
+
+        except Exception as e:
+            return ParseResult(error=f"Cable parse error: {str(e)}")
+
+    def _parse_raw_cable(self, content: str) -> ParseResult:
+        """Parse a raw-format diplomatic cable."""
+        metadata = {}
+
+        lines = content.split('\n')
+        body_start = 0
+
+        # Extract header fields
+        for i, line in enumerate(lines):
+            line_stripped = line.strip()
+            if not line_stripped:
+                continue
+
+            # Classification
+            for cls in CLASSIFICATIONS:
+                if line_stripped.startswith(cls):
+                    metadata['CLASSIFICATION'] = cls
+                    break
+
+            # Named fields
+            if line_stripped.startswith('SUBJECT:'):
+                metadata['SUBJECT'] = line_stripped[8:].strip()
+            elif line_stripped.startswith('TAGS:'):
+                metadata['TAGS'] = line_stripped[5:].strip()
+            elif line_stripped.startswith('REF:') or line_stripped.startswith('REFS:'):
+                ref_start = 4 if line_stripped.startswith('REF:') else 5
+                metadata['REFERENCE'] = line_stripped[ref_start:].strip()
+            elif line_stripped.startswith('E.O.'):
+                metadata['EXECUTIVE_ORDER'] = line_stripped[4:].strip()
+            elif line_stripped.startswith('FM '):
+                metadata['ORIGIN'] = line_stripped[3:].strip()
+            elif line_stripped.startswith('TO '):
+                metadata.setdefault('TO', []).append(line_stripped[3:].strip())
+
+            # Detect body start (after numbered paragraph or after header block)
+            if SECTION_PATTERN.match(line_stripped):
+                body_start = i
+                break
+            if line_stripped == 'BT' and i > 5:
+                body_start = i + 1
+                break
+
+        if body_start == 0:
+            # Fallback: body starts after a double blank line
+            for i in range(len(lines) - 1):
+                if not lines[i].strip() and not lines[i+1].strip():
+                    body_start = i + 2
+                    break
+
+        body = '\n'.join(lines[body_start:]).strip()
+
+        # Extract sections from body
+        sections = self._extract_sections(body)
+        if sections:
+            metadata['sections'] = sections
+
+        # Build formatted text
+        text_parts = []
+        for key in ('SUBJECT', 'ORIGIN', 'CLASSIFICATION', 'TAGS', 'REFERENCE'):
+            if key in metadata:
+                val = metadata[key]
+                if isinstance(val, list):
+                    val = ', '.join(val)
+                text_parts.append(f"{key}: {val}")
+
+        text_parts.append('')
+        text_parts.append(body)
+
+        text = '\n'.join(text_parts)
+        text = ANSI_ESCAPE.sub('', text)
+
+        return ParseResult(
+            success=True,
+            text=text,
+            metadata=metadata,
+            doc_type='cable',
+            title=metadata.get('SUBJECT', ''),
+            char_count=len(text),
+        )
+
+    def _parse_html_cable(self, content: str) -> ParseResult:
+        """Parse a WikiLeaks HTML-format cable page."""
+        try:
+            from bs4 import BeautifulSoup
+            soup = BeautifulSoup(content, 'html.parser')
+
+            metadata = {}
+
+            # WikiLeaks cable pages have specific CSS classes/structures
+            # Try common patterns
+            title_el = soup.find('title')
+            if title_el:
+                metadata['SUBJECT'] = title_el.get_text(strip=True)
+
+            # Look for cable metadata in page
+            for label in ('origin', 'classification', 'tags', 'subject', 'date'):
+                el = soup.find(class_=re.compile(label, re.I))
+                if el:
+                    metadata[label.upper()] = el.get_text(strip=True)
+
+            # Extract main content
+            body_el = (soup.find(id='cable-text') or
+                      soup.find(class_='cable-text') or
+                      soup.find('pre') or
+                      soup.find(class_='content'))
+
+            if body_el:
+                body = body_el.get_text(separator='\n', strip=True)
+            else:
+                body = soup.get_text(separator='\n', strip=True)
+
+            # Build text
+            text_parts = []
+            for key in ('SUBJECT', 'ORIGIN', 'CLASSIFICATION', 'TAGS'):
+                if key in metadata:
+                    text_parts.append(f"{key}: {metadata[key]}")
+            text_parts.append('')
+            text_parts.append(body)
+
+            text = '\n'.join(text_parts)
+            text = ANSI_ESCAPE.sub('', text)
+
+            return ParseResult(
+                success=True,
+                text=text,
+                metadata=metadata,
+                doc_type='cable',
+                title=metadata.get('SUBJECT', ''),
+                char_count=len(text),
+            )
+
+        except Exception as e:
+            return ParseResult(error=f"HTML cable parse error: {str(e)}")
+
+    def _extract_sections(self, body: str) -> list[dict]:
+        """Extract numbered sections from cable body."""
+        sections = []
+        current_num = None
+        current_text = []
+
+        for line in body.split('\n'):
+            match = SECTION_PATTERN.match(line.strip())
+            if match:
+                if current_num is not None:
+                    sections.append({
+                        'number': current_num,
+                        'text': '\n'.join(current_text).strip()
+                    })
+                # Extract section number and classification
+                current_num = line.strip().split('.')[0]
+                current_text = [line]
+            elif current_num is not None:
+                current_text.append(line)
+
+        if current_num is not None:
+            sections.append({
+                'number': current_num,
+                'text': '\n'.join(current_text).strip()
+            })
+
+        return sections
diff --git a/parsers/email_parser.py b/parsers/email_parser.py
new file mode 100644
index 0000000..830b978
--- /dev/null
+++ b/parsers/email_parser.py
@@ -0,0 +1,207 @@
+"""Email document parser for Mosaic — stdlib email parsing.
+
+SECURITY: Attachments are metadata-only. Never execute, open, or auto-extract
+attachment contents.
+"""
+import email
+import email.policy
+import re
+import hashlib
+import logging
+from typing import Optional
+
+from .base import ParseResult
+
+logger = logging.getLogger(__name__)
+
+# Suspicious MIME types that warrant flagging
+SUSPICIOUS_MIME_TYPES = {
+    'application/x-executable',
+    'application/x-msdos-program',
+    'application/x-msdownload',
+    'application/vnd.ms-excel.sheet.macroEnabled.12',
+    'application/vnd.ms-word.document.macroEnabled.12',
+    'application/x-shellscript',
+    'application/javascript',
+    'application/x-bat',
+    'application/x-com',
+    'application/x-dosexec',
+}
+
+# ANSI escape pattern
+ANSI_ESCAPE = re.compile(r'\x1b\[[0-9;]*[a-zA-Z]|\x1b\].*?\x07')
+
+
+class EmailDocParser:
+    """Parse email documents — headers + body, attachments as metadata only."""
+
+    def parse(self, file_path: str) -> ParseResult:
+        """Parse an email file (EML/MBOX format or HTML email page)."""
+        try:
+            with open(file_path, 'r', encoding='utf-8', errors='replace') as f:
+                content = f.read()
+
+            # Try standard email parsing first
+            msg = email.message_from_string(content, policy=email.policy.default)
+
+            # If it doesn't look like a real email, try HTML extraction
+            if not msg['From'] and not msg['Subject']:
+                return self._parse_html_email(content, file_path)
+
+            return self._parse_email_message(msg)
+
+        except Exception as e:
+            return ParseResult(error=f"Email parse error: {str(e)}")
+
+    def _parse_email_message(self, msg) -> ParseResult:
+        """Parse a standard email.Message object."""
+        # Extract headers
+        headers = {}
+        for key in ('From', 'To', 'Cc', 'Bcc', 'Subject', 'Date',
+                     'Message-ID', 'In-Reply-To', 'References'):
+            val = msg.get(key)
+            if val:
+                headers[key] = str(val)
+
+        # Extract body
+        body_parts = []
+        if msg.is_multipart():
+            for part in msg.walk():
+                content_type = part.get_content_type()
+                disposition = str(part.get('Content-Disposition', ''))
+
+                if 'attachment' in disposition:
+                    continue  # Skip attachments (metadata extracted separately)
+
+                if content_type == 'text/plain':
+                    payload = part.get_payload(decode=True)
+                    if payload:
+                        body_parts.append(payload.decode('utf-8', errors='replace'))
+                elif content_type == 'text/html':
+                    payload = part.get_payload(decode=True)
+                    if payload:
+                        # Simple HTML → text
+                        text = self._html_to_text(payload.decode('utf-8', errors='replace'))
+                        body_parts.append(text)
+        else:
+            payload = msg.get_payload(decode=True)
+            if payload:
+                body_parts.append(payload.decode('utf-8', errors='replace'))
+
+        body = '\n\n'.join(body_parts)
+
+        # Extract attachment metadata (NEVER the content)
+        attachments = self._extract_attachment_metadata(msg)
+
+        # Build formatted output
+        text_parts = []
+        for key, val in headers.items():
+            text_parts.append(f"{key}: {val}")
+        text_parts.append('')  # blank line
+        text_parts.append(body)
+
+        text = '\n'.join(text_parts)
+        # Strip ANSI
+        text = ANSI_ESCAPE.sub('', text)
+
+        metadata = {
+            'headers': headers,
+            'attachments': attachments,
+            'attachment_count': len(attachments),
+            'has_suspicious_attachments': any(
+                a.get('suspicious') for a in attachments
+            ),
+        }
+
+        return ParseResult(
+            success=True,
+            text=text,
+            metadata=metadata,
+            doc_type='email',
+            title=headers.get('Subject', ''),
+            char_count=len(text),
+        )
+
+    def _extract_attachment_metadata(self, msg) -> list[dict]:
+        """Extract attachment metadata ONLY — never content."""
+        attachments = []
+        if not msg.is_multipart():
+            return attachments
+
+        for part in msg.walk():
+            disposition = str(part.get('Content-Disposition', ''))
+            if 'attachment' not in disposition:
+                continue
+
+            filename = part.get_filename() or 'unnamed'
+            content_type = part.get_content_type()
+            payload = part.get_payload(decode=True)
+            size = len(payload) if payload else 0
+            md5 = hashlib.md5(payload).hexdigest() if payload else ''
+
+            att = {
+                'filename': filename,
+                'content_type': content_type,
+                'size': size,
+                'md5': md5,
+                'suspicious': content_type in SUSPICIOUS_MIME_TYPES,
+            }
+
+            if att['suspicious']:
+                logger.warning("Suspicious attachment: %s (%s)", filename, content_type)
+
+            attachments.append(att)
+
+        return attachments
+
+    def _parse_html_email(self, content: str, file_path: str) -> ParseResult:
+        """Parse an HTML page that contains email content (e.g., WikiLeaks email viewer)."""
+        try:
+            from bs4 import BeautifulSoup
+            soup = BeautifulSoup(content, 'html.parser')
+
+            # Try to extract email fields from HTML structure
+            metadata = {}
+            text_parts = []
+
+            # Look for common WikiLeaks email display patterns
+            for field in ('from', 'to', 'cc', 'subject', 'date'):
+                el = soup.find(class_=re.compile(field, re.I))
+                if el:
+                    metadata[field] = el.get_text(strip=True)
+                    text_parts.append(f"{field.title()}: {metadata[field]}")
+
+            # Get body
+            body = soup.find(class_=re.compile('body|content|message', re.I))
+            if body:
+                text_parts.append('')
+                text_parts.append(body.get_text(separator='\n', strip=True))
+            else:
+                # Fallback: get all text
+                text_parts.append('')
+                text_parts.append(soup.get_text(separator='\n', strip=True))
+
+            text = '\n'.join(text_parts)
+            text = ANSI_ESCAPE.sub('', text)
+
+            return ParseResult(
+                success=True,
+                text=text,
+                metadata={'headers': metadata, 'attachments': [], 'parsed_from_html': True},
+                doc_type='email',
+                title=metadata.get('subject', ''),
+                char_count=len(text),
+            )
+        except Exception as e:
+            return ParseResult(error=f"HTML email parse error: {str(e)}")
+
+    @staticmethod
+    def _html_to_text(html: str) -> str:
+        """Simple HTML to text conversion."""
+        try:
+            from bs4 import BeautifulSoup
+            soup = BeautifulSoup(html, 'html.parser')
+            return soup.get_text(separator='\n', strip=True)
+        except Exception:
+            # Fallback: strip tags with regex
+            return re.sub(r'<[^>]+>', '', html)
diff --git a/parsers/html_parser.py b/parsers/html_parser.py
new file mode 100644
index 0000000..cf182e0
--- /dev/null
+++ b/parsers/html_parser.py
@@ -0,0 +1,125 @@
+"""HTML document parser for Mosaic — BS4 with html.parser backend."""
+import re
+import logging
+
+from bs4 import BeautifulSoup
+
+from .base import ParseResult
+
+logger = logging.getLogger(__name__)
+
+# Elements to remove (navigation, scripts, styles)
+REMOVE_TAGS = ['script', 'style', 'nav', 'footer', 'header', 'aside',
+               'noscript', 'iframe', 'svg']
+
+# ANSI escape sequence pattern
+ANSI_ESCAPE = re.compile(r'\x1b\[[0-9;]*[a-zA-Z]|\x1b\].*?\x07')
+
+
+class HTMLDocParser:
+    """Parse HTML documents into structured text."""
+
+    def parse(self, file_path: str) -> ParseResult:
+        """Parse an HTML file.
+
+        Uses BeautifulSoup with 'html.parser' backend (explicitly specified,
+        not auto-detected) to extract structured text.
+        """
+        try:
+            with open(file_path, 'r', encoding='utf-8', errors='replace') as f:
+                raw_html = f.read()
+
+            soup = BeautifulSoup(raw_html, 'html.parser')
+
+            # Extract title
+            title = ""
+            title_tag = soup.find('title')
+            if title_tag:
+                title = title_tag.get_text(strip=True)
+
+            # Extract metadata
+            metadata = self._extract_metadata(soup)
+
+            # Remove unwanted elements
+            for tag_name in REMOVE_TAGS:
+                for tag in soup.find_all(tag_name):
+                    tag.decompose()
+
+            # Extract text preserving structure
+            text = self._extract_structured_text(soup)
+
+            # Strip ANSI escape sequences
+            text = ANSI_ESCAPE.sub('', text)
+
+            # Extract tables separately
+            tables = self._extract_tables(soup)
+            if tables:
+                metadata['tables'] = tables
+
+            char_count = len(text)
+
+            return ParseResult(
+                success=True,
+                text=text,
+                metadata=metadata,
+                doc_type='html',
+                title=title,
+                char_count=char_count,
+            )
+
+        except Exception as e:
+            return ParseResult(error=f"HTML parse error: {str(e)}")
+
+    def _extract_metadata(self, soup: BeautifulSoup) -> dict:
+        """Extract metadata from HTML meta tags."""
+        metadata = {}
+        for meta in soup.find_all('meta'):
+            name = meta.get('name', meta.get('property', ''))
+            content = meta.get('content', '')
+            if name and content:
+                metadata[name] = content
+        return metadata
+
+    def _extract_structured_text(self, soup: BeautifulSoup) -> str:
+        """Extract text preserving heading structure."""
+        parts = []
+
+        for element in soup.find_all(['h1', 'h2', 'h3', 'h4', 'h5', 'h6', 'p',
+                                       'li', 'td', 'th', 'pre', 'blockquote', 'div']):
+            text = element.get_text(separator=' ', strip=True)
+            if not text:
+                continue
+
+            tag = element.name
+            if tag.startswith('h'):
+                level = int(tag[1])
+                prefix = '#' * level
+                parts.append(f"\n{prefix} {text}\n")
+            elif tag == 'li':
+                parts.append(f"  - {text}")
+            elif tag == 'pre':
+                parts.append(f"\n```\n{text}\n```\n")
+            elif tag == 'blockquote':
+                parts.append(f"> {text}")
+            else:
+                parts.append(text)
+
+        result = '\n'.join(parts)
+        # Collapse excessive whitespace
+        result = re.sub(r'\n{3,}', '\n\n', result)
+        return result.strip()
+
+    def _extract_tables(self, soup: BeautifulSoup) -> list[dict]:
+        """Extract tables as structured data."""
+        tables = []
+        for table in soup.find_all('table'):
+            rows = []
+            for tr in table.find_all('tr'):
+                cells = []
+                for td in tr.find_all(['td', 'th']):
+                    cells.append(td.get_text(strip=True))
+                if cells:
+                    rows.append(cells)
+            if rows:
+                tables.append({'rows': rows})
+        return tables
diff --git a/parsers/pdf_parser.py b/parsers/pdf_parser.py
new file mode 100644
index 0000000..24c1890
--- /dev/null
+++ b/parsers/pdf_parser.py
@@ -0,0 +1,82 @@
+"""PDF document parser for Mosaic — PyMuPDF text extraction."""
+import logging
+from typing import Optional
+
+from .base import ParseResult
+
+logger = logging.getLogger(__name__)
+
+
+class PDFDocParser:
+    """Parse PDF documents using PyMuPDF (fitz) for text extraction."""
+
+    def __init__(self, max_pages: int = 200, min_chars_per_page: int = 50):
+        self.max_pages = max_pages
+        self.min_chars_per_page = min_chars_per_page
+
+    def parse(self, file_path: str) -> ParseResult:
+        """Parse a PDF file.
+
+        Extracts text page by page. Large docs (>max_pages) are split.
+        After extraction, checks chars/page ratio for OCR detection.
+        """
+        try:
+            import fitz  # PyMuPDF
+        except ImportError:
+            return ParseResult(error="PyMuPDF (fitz) not installed")
+
+        try:
+            doc = fitz.open(file_path)
+            total_pages = len(doc)
+            metadata = {
+                'page_count': total_pages,
+                'format': doc.metadata.get('format', ''),
+                'title': doc.metadata.get('title', ''),
+                'author': doc.metadata.get('author', ''),
+                'subject': doc.metadata.get('subject', ''),
+                'creator': doc.metadata.get('creator', ''),
+            }
+
+            title = metadata['title'] or ''
+
+            # Limit pages
+            pages_to_process = min(total_pages, self.max_pages)
+            if total_pages > self.max_pages:
+                logger.warning("PDF has %d pages, processing first %d",
+                             total_pages, self.max_pages)
+                metadata['truncated'] = True
+                metadata['truncated_at_page'] = self.max_pages
+
+            # Extract text page by page with form feed separators
+            text_parts = []
+            for page_num in range(pages_to_process):
+                page = doc[page_num]
+                page_text = page.get_text("text")
+                if page_text.strip():
+                    text_parts.append(page_text.strip())
+
+            doc.close()
+
+            text = '\f'.join(text_parts)  # Form feed between pages
+            char_count = len(text)
+
+            # Check for OCR need
+            needs_ocr = False
+            if pages_to_process > 0:
+                chars_per_page = char_count / pages_to_process
+                if chars_per_page < self.min_chars_per_page:
+                    needs_ocr = True
+                    metadata['chars_per_page'] = chars_per_page
+
+            return ParseResult(
+                success=True,
+                text=text,
+                metadata=metadata,
+                doc_type='pdf',
+                title=title,
+                char_count=char_count,
+                needs_ocr=needs_ocr,
+            )
+
+        except Exception as e:
+            return ParseResult(error=f"PDF parse error: {str(e)}")
diff --git a/parsers/text_parser.py b/parsers/text_parser.py
new file mode 100644
index 0000000..1a0300f
--- /dev/null
+++ b/parsers/text_parser.py
@@ -0,0 +1,102 @@
+"""Plain text / markdown parser for Mosaic."""
+import re
+import logging
+
+from .base import ParseResult
+
+logger = logging.getLogger(__name__)
+
+# ANSI escape pattern
+ANSI_ESCAPE = re.compile(r'\x1b\[[0-9;]*[a-zA-Z]|\x1b\].*?\x07')
+
+
+class TextDocParser:
+    """Parse plain text and markdown documents."""
+
+    def parse(self, file_path: str) -> ParseResult:
+        """Parse a plain text file."""
+        try:
+            # Try UTF-8 first, fall back to latin-1
+            try:
+                with open(file_path, 'r', encoding='utf-8') as f:
+                    text = f.read()
+            except UnicodeDecodeError:
+                with open(file_path, 'r', encoding='latin-1') as f:
+                    text = f.read()
+
+            # Strip ANSI escape sequences
+            text = ANSI_ESCAPE.sub('', text)
+
+            # Detect if it's markdown
+            is_markdown = self._detect_markdown(text, file_path)
+
+            # Extract basic metadata
+            metadata = {
+                'is_markdown': is_markdown,
+            }
+
+            # Try to extract a title
+            title = self._extract_title(text)
+
+            # Basic section detection
+            sections = self._detect_sections(text)
+            if sections:
+                metadata['section_count'] = len(sections)
+
+            return ParseResult(
+                success=True,
+                text=text.strip(),
+                metadata=metadata,
+                doc_type='text',
+                title=title,
+                char_count=len(text),
+            )
+
+        except Exception as e:
+            return ParseResult(error=f"Text parse error: {str(e)}")
+
+    def _detect_markdown(self, text: str, file_path: str) -> bool:
+        """Detect if text is markdown."""
+        if file_path.endswith(('.md', '.markdown', '.mkd')):
+            return True
+        # Check for markdown indicators
+        md_patterns = [
+            r'^#{1,6}\s',      # Headers
+            r'^\*\*[^*]+\*\*', # Bold
+            r'^\- ',           # List items
+            r'^\d+\. ',       # Numbered list
+            r'\[.+\]\(.+\)',  # Links
+        ]
+        matches = sum(1 for p in md_patterns if re.search(p, text[:2000], re.MULTILINE))
+        return matches >= 2
+
+    def _extract_title(self, text: str) -> str:
+        """Extract title from first heading or first line."""
+        lines = text.strip().split('\n')
+        for line in lines[:5]:
+            line = line.strip()
+            if not line:
+                continue
+            # Markdown heading
+            match = re.match(r'^#{1,3}\s+(.+)', line)
+            if match:
+                return match.group(1).strip()
+            # First non-empty line
+            if len(line) < 200:
+                return line
+            break
+        return ""
+
+    def _detect_sections(self, text: str) -> list[str]:
+        """Detect section headings in text."""
+        sections = []
+        for line in text.split('\n'):
+            line = line.strip()
+            # Markdown headings
+            if re.match(r'^#{1,6}\s+', line):
+                sections.append(line)
+            # All-caps headings
+            elif (line.isupper() and 3 < len(line) < 100 and
+                  not line.startswith(('---', '===', '***'))):
+                sections.append(line)
+        return sections