Merge pull request #1617 from CREDO23/feature-okf

[Feat] Serve the knowledge base in Open Knowledge Format (OKF)

surfsense_backend/app/routes/documents_routes.py

@@ -1,7 +1,8 @@
 # Force asyncio to use standard event loop before unstructured imports
 import asyncio
 
-from fastapi import APIRouter, Depends, Form, HTTPException, Query, UploadFile
+from fastapi import APIRouter, Depends, Form, HTTPException, Query, Request, UploadFile
+from fastapi.responses import PlainTextResponse
 from pydantic import BaseModel as PydanticBaseModel, Field
 from sqlalchemy.ext.asyncio import AsyncSession
 from sqlalchemy.future import select
@@ -34,6 +35,8 @@ from app.schemas import (
     FolderRead,
     PaginatedResponse,
 )
+from app.services.export_service import resolve_document_markdown
+from app.services.okf import document_to_concept
 from app.services.task_dispatcher import TaskDispatcher, get_task_dispatcher
 from app.users import get_auth_context
 from app.utils.rbac import check_permission
@@ -1255,12 +1258,14 @@ async def get_document_chunks_paginated(
 @router.get("/documents/{document_id}", response_model=DocumentRead)
 async def read_document(
     document_id: int,
+    request: Request,
     session: AsyncSession = Depends(get_async_session),
     auth: AuthContext = Depends(get_auth_context),
 ):
-    """
-    Get a specific document by ID.
-    Requires DOCUMENTS_READ permission for the workspace.
+    """Get a document by ID.
+
+    ``Accept: text/markdown`` returns the OKF concept; otherwise the JSON record.
+    Requires DOCUMENTS_READ permission.
     """
     try:
         result = await session.execute(
@@ -1282,6 +1287,16 @@ async def read_document(
             "You don't have permission to read documents in this workspace",
         )
 
+        # ponytail: substring match, not RFC 7231 q-values (OKF is the only non-JSON view).
+        if "text/markdown" in request.headers.get("accept", ""):
+            body = (
+                await resolve_document_markdown(session, document)
+                or document.content
+                or ""
+            )
+            concept = document_to_concept(document, body=body)
+            return PlainTextResponse(concept, media_type="text/markdown")
+
         raw_content = document.content or ""
         return DocumentRead(
             id=document.id,

surfsense_backend/app/services/export_service.py

@@ -12,9 +12,26 @@ from sqlalchemy.future import select
 
 from app.db import Chunk, Document, Folder
 from app.services.folder_service import get_folder_subtree_ids
+from app.services.okf import (
+    INDEX_FILENAME,
+    LOG_FILENAME,
+    ConceptRef,
+    LogEntry,
+    SubdirRef,
+    document_to_concept,
+    folder_to_index,
+    folder_to_log,
+    okf_type,
+)
 
 logger = logging.getLogger(__name__)
 
+# Root index.md declares the targeted OKF version in frontmatter - the one place
+# the spec permits frontmatter in an index file.
+_ROOT_INDEX_FRONTMATTER = '---\nokf_version: "0.1"\n---\n\n'
+
+_RESERVED_STEMS = {"index", "log"}
+
 
 def _sanitize_filename(title: str) -> str:
     safe = "".join(c if c.isalnum() or c in " -_." else "_" for c in title).strip()
@@ -43,7 +60,7 @@ def _build_folder_path_map(folders: list[Folder]) -> dict[int, str]:
     return cache
 
 
-async def _get_document_markdown(
+async def resolve_document_markdown(
     session: AsyncSession, document: Document
 ) -> str | None:
     """Resolve markdown content using the 3-tier fallback:
@@ -71,6 +88,67 @@ async def _get_document_markdown(
     return None
 
 
+def _build_index_files(
+    dir_concepts: dict[str, list[ConceptRef]],
+) -> list[tuple[str, str]]:
+    """Build ``index.md`` files for every directory (and ancestor) with content.
+
+    Produces OKF progressive-disclosure listings: each directory lists its
+    concepts (grouped by type) and its immediate subdirectories. The bundle-root
+    index also declares ``okf_version``.
+    """
+    all_dirs: set[str] = {""}
+    for dir_path in dir_concepts:
+        all_dirs.add(dir_path)
+        parts = dir_path.split("/") if dir_path else []
+        for i in range(1, len(parts)):
+            all_dirs.add("/".join(parts[:i]))
+
+    children_by_dir: dict[str, list[str]] = {}
+    for dir_path in all_dirs:
+        if not dir_path:
+            continue
+        parent = dir_path.rsplit("/", 1)[0] if "/" in dir_path else ""
+        children_by_dir.setdefault(parent, []).append(dir_path)
+
+    index_files: list[tuple[str, str]] = []
+    for dir_path in all_dirs:
+        subdirs = [
+            SubdirRef(name=child.rsplit("/", 1)[-1])
+            for child in children_by_dir.get(dir_path, [])
+        ]
+        body = folder_to_index(
+            concepts=dir_concepts.get(dir_path, []),
+            subdirectories=subdirs,
+        )
+        if not body:
+            continue
+        if dir_path:
+            index_files.append((f"{dir_path}/{INDEX_FILENAME}", body))
+        else:
+            index_files.append((INDEX_FILENAME, _ROOT_INDEX_FRONTMATTER + body))
+
+    return index_files
+
+
+def _build_log_files(
+    dir_logs: dict[str, list[LogEntry]],
+) -> list[tuple[str, str]]:
+    """Build ``log.md`` files for every directory that holds concepts.
+
+    Unlike ``index.md``, logs are only synthesized where documents actually live
+    (no ancestor synthesis): an empty intermediate directory has nothing to log.
+    """
+    log_files: list[tuple[str, str]] = []
+    for dir_path, entries in dir_logs.items():
+        body = folder_to_log(entries)
+        if not body:
+            continue
+        path = f"{dir_path}/{LOG_FILENAME}" if dir_path else LOG_FILENAME
+        log_files.append((path, body))
+    return log_files
+
+
 @dataclass
 class ExportResult:
     zip_path: str
@@ -120,6 +198,10 @@ async def build_export_zip(
     used_paths: dict[str, int] = {}
     skipped_docs: list[str] = []
     is_first_batch = True
+    # dir path -> concepts it holds, accumulated across batches for index.md.
+    dir_concepts: dict[str, list[ConceptRef]] = {}
+    # dir path -> log entries it holds, accumulated across batches for log.md.
+    dir_logs: dict[str, list[LogEntry]] = {}
 
     try:
         offset = 0
@@ -143,7 +225,7 @@ async def build_export_zip(
                     skipped_docs.append(doc.title or "Untitled")
                     continue
 
-                markdown = await _get_document_markdown(session, doc)
+                markdown = await resolve_document_markdown(session, doc)
                 if not markdown or not markdown.strip():
                     continue
 
@@ -153,6 +235,9 @@ async def build_export_zip(
                     dir_path = ""
 
                 base_name = _sanitize_filename(doc.title or "Untitled")
+                # Never collide with reserved OKF filenames (index.md, log.md).
+                if base_name.lower() in _RESERVED_STEMS:
+                    base_name = f"{base_name}_"
                 file_path = (
                     f"{dir_path}/{base_name}.md" if dir_path else f"{base_name}.md"
                 )
@@ -160,14 +245,41 @@ async def build_export_zip(
                 if file_path in used_paths:
                     used_paths[file_path] += 1
                     suffix = used_paths[file_path]
+                    base_name = f"{base_name}_{suffix}"
                     file_path = (
-                        f"{dir_path}/{base_name}_{suffix}.md"
+                        f"{dir_path}/{base_name}.md"
                         if dir_path
-                        else f"{base_name}_{suffix}.md"
+                        else f"{base_name}.md"
                     )
                 used_paths[file_path] = used_paths.get(file_path, 0) + 1
 
-                entries.append((file_path, markdown))
+                concept = document_to_concept(doc, body=markdown)
+                entries.append((file_path, concept))
+
+                metadata = (
+                    doc.document_metadata
+                    if isinstance(doc.document_metadata, dict)
+                    else {}
+                )
+                description = metadata.get("description")
+                dir_concepts.setdefault(dir_path, []).append(
+                    ConceptRef(
+                        title=doc.title or "Untitled",
+                        filename=f"{base_name}.md",
+                        type=okf_type(doc.document_type),
+                        description=description
+                        if isinstance(description, str) and description.strip()
+                        else None,
+                    )
+                )
+
+                changed_at = doc.updated_at or doc.created_at
+                dir_logs.setdefault(dir_path, []).append(
+                    LogEntry(
+                        title=doc.title or "Untitled",
+                        timestamp=changed_at.isoformat() if changed_at else None,
+                    )
+                )
 
             if entries:
                 mode = "w" if is_first_batch else "a"
@@ -183,6 +295,30 @@ async def build_export_zip(
 
             offset += batch_size
 
+        index_files = _build_index_files(dir_concepts)
+        if index_files:
+            mode = "w" if is_first_batch else "a"
+
+            def _write_indexes(m: str = mode, e: list = index_files) -> None:
+                with zipfile.ZipFile(tmp_path, m, zipfile.ZIP_DEFLATED) as zf:
+                    for path, content in e:
+                        zf.writestr(path, content)
+
+            await asyncio.to_thread(_write_indexes)
+            is_first_batch = False
+
+        log_files = _build_log_files(dir_logs)
+        if log_files:
+            mode = "w" if is_first_batch else "a"
+
+            def _write_logs(m: str = mode, e: list = log_files) -> None:
+                with zipfile.ZipFile(tmp_path, m, zipfile.ZIP_DEFLATED) as zf:
+                    for path, content in e:
+                        zf.writestr(path, content)
+
+            await asyncio.to_thread(_write_logs)
+            is_first_batch = False
+
         export_name = "knowledge-base"
         if folder_id is not None and folder_id in folder_path_map:
             export_name = _sanitize_filename(folder_path_map[folder_id].split("/")[0])

surfsense_backend/app/services/okf/__init__.py

@@ -0,0 +1,56 @@
+"""Open Knowledge Format (OKF v0.1) serialization for the SurfSense KB.
+
+Single source of truth for turning documents and folders into OKF concepts,
+``index.md`` listings and ``log.md`` logs. Pure KB-layer functions; every
+consumer (ZIP export, REST, MCP, agents) calls in.
+
+The OKF-native model: ``Document`` rows are canonical, and frontmatter is
+*derived* from their columns on read (never stored), so rows are conformant by
+construction. Chunks and embeddings are a *derived*, rebuildable search
+projection - never a source of truth.
+
+Spec: https://github.com/GoogleCloudPlatform/knowledge-catalog/blob/main/okf/SPEC.md
+"""
+
+from app.services.okf.serializer import (
+    INDEX_FILENAME,
+    LOG_FILENAME,
+    ConceptRef,
+    LogEntry,
+    SubdirRef,
+    build_frontmatter,
+    document_to_concept,
+    folder_to_index,
+    folder_to_log,
+    render_frontmatter,
+)
+from app.services.okf.type_mapping import okf_resource, okf_type
+from app.services.okf.validator import (
+    RECOMMENDED_FRONTMATTER_KEYS,
+    REQUIRED_FRONTMATTER_KEYS,
+    is_conformant_concept,
+    parse_frontmatter,
+    validate_bundle,
+    validate_concept,
+)
+
+__all__ = [
+    "INDEX_FILENAME",
+    "LOG_FILENAME",
+    "ConceptRef",
+    "LogEntry",
+    "SubdirRef",
+    "build_frontmatter",
+    "document_to_concept",
+    "folder_to_index",
+    "folder_to_log",
+    "render_frontmatter",
+    "okf_resource",
+    "okf_type",
+    "RECOMMENDED_FRONTMATTER_KEYS",
+    "REQUIRED_FRONTMATTER_KEYS",
+    "is_conformant_concept",
+    "parse_frontmatter",
+    "validate_bundle",
+    "validate_concept",
+]

surfsense_backend/app/services/okf/serializer.py

@@ -0,0 +1,193 @@
+"""Serialize SurfSense knowledge into Open Knowledge Format (OKF v0.1).
+
+Pure functions with no HTTP / MCP / framework dependencies: given a
+:class:`~app.db.Document` (and, for listings, its neighbours) they return
+OKF-conformant markdown. Every consumer (ZIP export, REST, MCP, agents) calls
+these rather than re-implementing frontmatter.
+
+Spec: https://github.com/GoogleCloudPlatform/knowledge-catalog/blob/main/okf/SPEC.md
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Any
+
+import yaml
+
+from app.db import Document
+from app.services.okf.type_mapping import okf_resource, okf_type
+
+# Reserved OKF filenames; never used for concept documents.
+INDEX_FILENAME = "index.md"
+LOG_FILENAME = "log.md"
+
+_FRONTMATTER_DELIMITER = "---"
+
+
+def _tags_from_metadata(metadata: dict[str, Any] | None) -> list[str] | None:
+    if not isinstance(metadata, dict):
+        return None
+    raw = metadata.get("tags")
+    if isinstance(raw, list):
+        tags = [str(t).strip() for t in raw if str(t).strip()]
+        return tags or None
+    return None
+
+
+def _timestamp(document: Document) -> str | None:
+    when = document.updated_at or document.created_at
+    if when is None:
+        return None
+    # ISO 8601, matching Google's sample bundles (e.g. 2026-05-28T22:49:59+00:00).
+    return when.isoformat()
+
+
+def build_frontmatter(document: Document) -> dict[str, Any]:
+    """Build the ordered OKF frontmatter mapping for a document.
+
+    Only ``type`` is required; recommended keys are included only when we have a
+    value. Insertion order is preserved in the emitted YAML.
+    """
+    metadata = document.document_metadata if isinstance(document.document_metadata, dict) else {}
+
+    frontmatter: dict[str, Any] = {"type": okf_type(document.document_type)}
+
+    resource = okf_resource(document.document_type, metadata)
+    if resource:
+        frontmatter["resource"] = resource
+
+    title = (document.title or "").strip()
+    if title:
+        frontmatter["title"] = title
+
+    description = metadata.get("description")
+    if isinstance(description, str) and description.strip():
+        frontmatter["description"] = description.strip()
+
+    tags = _tags_from_metadata(metadata)
+    if tags:
+        frontmatter["tags"] = tags
+
+    timestamp = _timestamp(document)
+    if timestamp:
+        frontmatter["timestamp"] = timestamp
+
+    return frontmatter
+
+
+def render_frontmatter(frontmatter: dict[str, Any]) -> str:
+    """Render a frontmatter mapping as a YAML block delimited by ``---``."""
+    body = yaml.safe_dump(
+        frontmatter,
+        sort_keys=False,
+        allow_unicode=True,
+        default_flow_style=False,
+    )
+    return f"{_FRONTMATTER_DELIMITER}\n{body}{_FRONTMATTER_DELIMITER}\n"
+
+
+def document_to_concept(document: Document, *, body: str) -> str:
+    """Serialize a document as an OKF concept: frontmatter + ``body``.
+
+    ``body`` is caller-resolved markdown, keeping this a pure formatting step.
+    """
+    frontmatter = render_frontmatter(build_frontmatter(document))
+    body_text = (body or "").strip("\n")
+    return f"{frontmatter}\n{body_text}\n"
+
+
+@dataclass(frozen=True)
+class ConceptRef:
+    """One concept entry for an ``index.md`` listing."""
+
+    title: str
+    filename: str  # relative to the directory, e.g. "orders.md"
+    type: str  # OKF type, used as the grouping heading
+    description: str | None = None
+
+
+@dataclass(frozen=True)
+class SubdirRef:
+    """One subdirectory entry for an ``index.md`` listing."""
+
+    name: str  # directory name, e.g. "tables"
+    description: str | None = None
+
+
+@dataclass(frozen=True)
+class LogEntry:
+    """One line of an OKF ``log.md``: a concept and when it last changed."""
+
+    title: str
+    timestamp: str | None = None  # ISO-8601, or None when unknown
+
+
+def folder_to_log(entries: list[LogEntry]) -> str:
+    """Build a minimal OKF ``log.md`` body for one directory.
+
+    Lists each concept newest-first with the time it last changed (from a
+    document's ``updated_at``/``created_at``); undated entries sort last. Returns
+    an empty string when there is nothing to log.
+
+    ponytail: this is a last-touched summary, not a per-field change history. A
+    fuller changelog would read ``DocumentVersion`` rows; upgrade there if
+    consumers ever need diffs rather than "what changed and when".
+    """
+    if not entries:
+        return ""
+    ordered = sorted(
+        entries,
+        key=lambda e: (e.timestamp is not None, e.timestamp or "", e.title),
+        reverse=True,
+    )
+    lines = ["# Change Log", ""]
+    for entry in ordered:
+        when = f" - {entry.timestamp}" if entry.timestamp else ""
+        lines.append(f"* {entry.title}{when}")
+    return "\n".join(lines) + "\n"
+
+
+def _index_bullet(title: str, link: str, description: str | None) -> str:
+    bullet = f"* [{title}]({link})"
+    if description:
+        # Keep index descriptions to a single line.
+        bullet += f" - {' '.join(description.split())}"
+    return bullet
+
+
+def folder_to_index(
+    *,
+    concepts: list[ConceptRef] | None = None,
+    subdirectories: list[SubdirRef] | None = None,
+) -> str:
+    """Build an OKF ``index.md`` body (no frontmatter) for one directory.
+
+    Subdirectories are listed under a ``# Subdirectories`` heading and concepts
+    are grouped under their ``type`` heading, mirroring Google's sample bundles.
+    Returns an empty string when there is nothing to list.
+    """
+    concepts = concepts or []
+    subdirectories = subdirectories or []
+    sections: list[str] = []
+
+    if subdirectories:
+        lines = ["# Subdirectories", ""]
+        for sub in sorted(subdirectories, key=lambda s: s.name.lower()):
+            lines.append(
+                _index_bullet(sub.name, f"{sub.name}/{INDEX_FILENAME}", sub.description)
+            )
+        sections.append("\n".join(lines))
+
+    by_type: dict[str, list[ConceptRef]] = {}
+    for concept in concepts:
+        by_type.setdefault(concept.type, []).append(concept)
+    for type_heading in sorted(by_type):
+        lines = [f"# {type_heading}", ""]
+        for concept in sorted(by_type[type_heading], key=lambda c: c.title.lower()):
+            lines.append(
+                _index_bullet(concept.title, concept.filename, concept.description)
+            )
+        sections.append("\n".join(lines))
+
+    return ("\n\n".join(sections) + "\n") if sections else ""

surfsense_backend/app/services/okf/type_mapping.py

@@ -0,0 +1,124 @@
+"""Map SurfSense document types to OKF concept ``type`` strings and ``resource`` URIs.
+
+OKF (Open Knowledge Format v0.1) requires a non-empty, human-friendly ``type`` on
+every concept and recommends a ``resource`` URI pointing at the underlying asset.
+SurfSense stores the type as a :class:`~app.db.DocumentType` enum and the source
+URL (when one exists) inside the free-form ``document_metadata`` JSON, under keys
+that differ per connector. This module is the single place that knows both mappings.
+
+Spec: https://github.com/GoogleCloudPlatform/knowledge-catalog/blob/main/okf/SPEC.md
+"""
+
+from __future__ import annotations
+
+from typing import Any
+
+from app.db import DocumentType
+
+# Title-cased type strings per Google's sample bundles. Types absent here fall
+# back to a derived string (see ``okf_type``), so this need not be exhaustive.
+OKF_TYPE_BY_DOCUMENT_TYPE: dict[DocumentType, str] = {
+    DocumentType.EXTENSION: "Web Page",
+    DocumentType.CRAWLED_URL: "Web Page",
+    DocumentType.FILE: "File",
+    DocumentType.NOTE: "Note",
+    DocumentType.SLACK_CONNECTOR: "Slack Message",
+    DocumentType.TEAMS_CONNECTOR: "Teams Message",
+    DocumentType.ONEDRIVE_FILE: "OneDrive File",
+    DocumentType.NOTION_CONNECTOR: "Notion Page",
+    DocumentType.YOUTUBE_VIDEO: "YouTube Video",
+    DocumentType.GITHUB_CONNECTOR: "GitHub Document",
+    DocumentType.LINEAR_CONNECTOR: "Linear Issue",
+    DocumentType.DISCORD_CONNECTOR: "Discord Message",
+    DocumentType.JIRA_CONNECTOR: "Jira Issue",
+    DocumentType.CONFLUENCE_CONNECTOR: "Confluence Page",
+    DocumentType.CLICKUP_CONNECTOR: "ClickUp Task",
+    DocumentType.GOOGLE_CALENDAR_CONNECTOR: "Google Calendar Event",
+    DocumentType.GOOGLE_GMAIL_CONNECTOR: "Gmail Message",
+    DocumentType.GOOGLE_DRIVE_FILE: "Google Drive File",
+    DocumentType.AIRTABLE_CONNECTOR: "Airtable Record",
+    DocumentType.LUMA_CONNECTOR: "Luma Event",
+    DocumentType.ELASTICSEARCH_CONNECTOR: "Elasticsearch Document",
+    DocumentType.BOOKSTACK_CONNECTOR: "BookStack Page",
+    DocumentType.CIRCLEBACK: "Circleback Meeting",
+    DocumentType.OBSIDIAN_CONNECTOR: "Obsidian Note",
+    DocumentType.DROPBOX_FILE: "Dropbox File",
+    DocumentType.COMPOSIO_GOOGLE_DRIVE_CONNECTOR: "Google Drive File",
+    DocumentType.COMPOSIO_GMAIL_CONNECTOR: "Gmail Message",
+    DocumentType.COMPOSIO_GOOGLE_CALENDAR_CONNECTOR: "Google Calendar Event",
+    DocumentType.LOCAL_FOLDER_FILE: "File",
+}
+
+
+def _coerce_document_type(document_type: DocumentType | str | None) -> DocumentType | None:
+    if document_type is None:
+        return None
+    try:
+        return DocumentType(document_type)
+    except ValueError:
+        return None
+
+
+def okf_type(document_type: DocumentType | str | None) -> str:
+    """Return the OKF ``type`` string for a SurfSense document type.
+
+    Falls back to a Title-Cased version of the enum value so the required
+    ``type`` field is never empty, even for types added after this map.
+    """
+    dt = _coerce_document_type(document_type)
+    if dt is None:
+        if document_type is None:
+            return "Document"
+        return str(document_type).replace("_", " ").title() or "Document"
+    return OKF_TYPE_BY_DOCUMENT_TYPE.get(dt) or dt.value.replace("_", " ").title()
+
+
+# Per-type metadata keys that hold the canonical source URL, checked first.
+_RESOURCE_KEYS_BY_TYPE: dict[DocumentType, tuple[str, ...]] = {
+    DocumentType.GOOGLE_DRIVE_FILE: ("webViewLink", "web_view_link"),
+    DocumentType.COMPOSIO_GOOGLE_DRIVE_CONNECTOR: ("webViewLink", "web_view_link"),
+    DocumentType.GITHUB_CONNECTOR: ("html_url", "url"),
+    DocumentType.NOTION_CONNECTOR: ("url",),
+    DocumentType.LINEAR_CONNECTOR: ("url",),
+    DocumentType.JIRA_CONNECTOR: ("url",),
+    DocumentType.CONFLUENCE_CONNECTOR: ("url",),
+    DocumentType.SLACK_CONNECTOR: ("permalink", "url"),
+    DocumentType.DISCORD_CONNECTOR: ("url",),
+    DocumentType.CLICKUP_CONNECTOR: ("url",),
+    DocumentType.YOUTUBE_VIDEO: ("url", "video_url"),
+    DocumentType.LUMA_CONNECTOR: ("url",),
+    DocumentType.BOOKSTACK_CONNECTOR: ("url",),
+}
+
+# Generic URL-bearing keys checked for any type as a fallback.
+_GENERIC_RESOURCE_KEYS: tuple[str, ...] = (
+    "url",
+    "URL",
+    "source_url",
+    "sourceUrl",
+    "source",
+    "link",
+    "permalink",
+    "web_url",
+    "webViewLink",
+    "html_url",
+)
+
+
+def okf_resource(
+    document_type: DocumentType | str | None, metadata: dict[str, Any] | None
+) -> str | None:
+    """Return the canonical source URI for a document, or ``None`` if it has none."""
+    if not isinstance(metadata, dict) or not metadata:
+        return None
+
+    dt = _coerce_document_type(document_type)
+    type_keys = _RESOURCE_KEYS_BY_TYPE.get(dt, ()) if dt is not None else ()
+
+    for key in (*type_keys, *_GENERIC_RESOURCE_KEYS):
+        value = metadata.get(key)
+        if isinstance(value, str):
+            candidate = value.strip()
+            if candidate.startswith(("http://", "https://")):
+                return candidate
+    return None

surfsense_backend/app/services/okf/validator.py

@@ -0,0 +1,90 @@
+"""Minimal OKF v0.1 conformance checks.
+
+A bundle conforms if every non-reserved ``.md`` file has a parseable YAML
+frontmatter block whose ``type`` is a non-empty string. Reserved files
+(``index.md``, ``log.md``) are exempt from the frontmatter requirement.
+Consumers must stay permissive, so this is for producers to self-check what they
+emit, not to reject foreign bundles.
+
+Spec: https://github.com/GoogleCloudPlatform/knowledge-catalog/blob/main/okf/SPEC.md
+"""
+
+from __future__ import annotations
+
+import yaml
+
+from app.services.okf.serializer import INDEX_FILENAME, LOG_FILENAME
+
+_DELIMITER = "---"
+RESERVED_FILENAMES = frozenset({INDEX_FILENAME, LOG_FILENAME})
+
+# The OKF v0.1 frontmatter contract, shared by serializer and validator. Only
+# ``type`` is mandatory; the rest are recommended.
+REQUIRED_FRONTMATTER_KEYS: tuple[str, ...] = ("type",)
+RECOMMENDED_FRONTMATTER_KEYS: tuple[str, ...] = (
+    "title",
+    "description",
+    "resource",
+    "tags",
+    "timestamp",
+)
+
+
+def parse_frontmatter(text: str) -> tuple[dict | None, str | None]:
+    """Split an OKF concept into ``(frontmatter_dict, error)``.
+
+    Returns ``(dict, None)`` on success or ``(None, reason)`` if the frontmatter
+    block is missing or not a parseable YAML mapping.
+    """
+    if not text.startswith(_DELIMITER + "\n"):
+        return None, "missing opening '---' frontmatter delimiter"
+    rest = text[len(_DELIMITER) + 1 :]
+    end = rest.find("\n" + _DELIMITER)
+    if end == -1:
+        return None, "missing closing '---' frontmatter delimiter"
+    block = rest[:end]
+    try:
+        data = yaml.safe_load(block)
+    except yaml.YAMLError as exc:
+        return None, f"frontmatter is not valid YAML: {exc}"
+    if not isinstance(data, dict):
+        return None, "frontmatter is not a YAML mapping"
+    return data, None
+
+
+def validate_concept(text: str) -> list[str]:
+    """Return conformance errors for a single concept document (empty == conforms)."""
+    frontmatter, error = parse_frontmatter(text)
+    if error:
+        return [error]
+    errors: list[str] = []
+    for key in REQUIRED_FRONTMATTER_KEYS:
+        value = frontmatter.get(key)
+        if not isinstance(value, str) or not value.strip():
+            errors.append(f"frontmatter '{key}' is missing or empty")
+    return errors
+
+
+def is_conformant_concept(text: str) -> bool:
+    """True if a single concept document conforms to OKF v0.1."""
+    return not validate_concept(text)
+
+
+def validate_bundle(files: dict[str, str]) -> dict[str, list[str]]:
+    """Validate a bundle given a mapping of relative path -> file contents.
+
+    Only non-reserved ``.md`` files are checked for concept conformance. Returns
+    a mapping of path -> errors for every non-conformant file (an empty mapping
+    means the whole bundle conforms).
+    """
+    problems: dict[str, list[str]] = {}
+    for path, contents in files.items():
+        if not path.endswith(".md"):
+            continue
+        filename = path.rsplit("/", 1)[-1]
+        if filename in RESERVED_FILENAMES:
+            continue
+        errors = validate_concept(contents)
+        if errors:
+            problems[path] = errors
+    return problems