keep-skill 0.1.0__py3-none-any.whl → 0.2.0__py3-none-any.whl

keep/api.py

@@ -8,17 +8,104 @@ This is the minimal working implementation focused on:
 - get(): retrieve by ID
 """
 
+import hashlib
+import logging
 import re
-from datetime import datetime, timezone
+from datetime import datetime, timezone, timedelta
 from pathlib import Path
 from typing import Any, Optional
 
+logger = logging.getLogger(__name__)
+
+
+def _parse_since(since: str) -> str:
+    """
+    Parse a 'since' string and return a YYYY-MM-DD cutoff date.
+
+    Accepts:
+    - ISO 8601 duration: P3D (3 days), P1W (1 week), PT1H (1 hour), P1DT12H, etc.
+    - ISO date: 2026-01-15
+    - Date with slashes: 2026/01/15
+
+    Returns:
+        YYYY-MM-DD string for the cutoff date
+    """
+    since = since.strip()
+
+    # ISO 8601 duration: P[n]Y[n]M[n]W[n]DT[n]H[n]M[n]S
+    if since.upper().startswith("P"):
+        duration_str = since.upper()
+
+        # Parse duration components
+        years = months = weeks = days = hours = minutes = seconds = 0
+
+        # Split on T to separate date and time parts
+        if "T" in duration_str:
+            date_part, time_part = duration_str.split("T", 1)
+        else:
+            date_part = duration_str
+            time_part = ""
+
+        # Parse date part (P[n]Y[n]M[n]W[n]D)
+        date_part = date_part[1:]  # Remove leading P
+        for match in re.finditer(r"(\d+)([YMWD])", date_part):
+            value, unit = int(match.group(1)), match.group(2)
+            if unit == "Y":
+                years = value
+            elif unit == "M":
+                months = value
+            elif unit == "W":
+                weeks = value
+            elif unit == "D":
+                days = value
+
+        # Parse time part ([n]H[n]M[n]S)
+        for match in re.finditer(r"(\d+)([HMS])", time_part):
+            value, unit = int(match.group(1)), match.group(2)
+            if unit == "H":
+                hours = value
+            elif unit == "M":
+                minutes = value
+            elif unit == "S":
+                seconds = value
+
+        # Convert to timedelta (approximate months/years)
+        total_days = years * 365 + months * 30 + weeks * 7 + days
+        delta = timedelta(days=total_days, hours=hours, minutes=minutes, seconds=seconds)
+        cutoff = datetime.now(timezone.utc) - delta
+        return cutoff.strftime("%Y-%m-%d")
+
+    # Try parsing as date
+    # ISO format: 2026-01-15 or 2026-01-15T...
+    # Slash format: 2026/01/15
+    date_str = since.replace("/", "-").split("T")[0]
+
+    try:
+        parsed = datetime.strptime(date_str, "%Y-%m-%d")
+        return parsed.strftime("%Y-%m-%d")
+    except ValueError:
+        pass
+
+    raise ValueError(
+        f"Invalid 'since' format: {since}. "
+        "Use ISO duration (P3D, PT1H, P1W) or date (2026-01-15)"
+    )
+
+
+def _filter_by_date(items: list, since: str) -> list:
+    """Filter items to only those updated since the given date/duration."""
+    cutoff = _parse_since(since)
+    return [
+        item for item in items
+        if item.tags.get("_updated_date", "0000-00-00") >= cutoff
+    ]
+
 import os
 import subprocess
 import sys
 
-from .config import load_or_create_config, StoreConfig
-from .paths import get_default_store_path
+from .config import load_or_create_config, save_config, StoreConfig, EmbeddingIdentity
+from .paths import get_config_dir, get_default_store_path
 from .pending_summaries import PendingSummaryQueue
 from .providers import get_registry
 from .providers.base import (
@@ -28,7 +115,7 @@ from .providers.base import (
 )
 from .providers.embedding_cache import CachingEmbeddingProvider
 from .store import ChromaStore
-from .types import Item, filter_non_system_tags
+from .types import Item, filter_non_system_tags, SYSTEM_TAG_PREFIX
 
 
 # Default max length for truncated placeholder summaries
@@ -41,6 +128,69 @@ MAX_SUMMARY_ATTEMPTS = 5
 # Collection name validation: lowercase ASCII and underscores only
 COLLECTION_NAME_PATTERN = re.compile(r"^[a-z][a-z0-9_]*$")
 
+# Environment variable prefix for auto-applied tags
+ENV_TAG_PREFIX = "KEEP_TAG_"
+
+# Fixed ID for the current working context (singleton)
+NOWDOC_ID = "_now:default"
+
+# Path to system documents
+SYSTEM_DOC_DIR = Path(__file__).parent.parent / "docs" / "system"
+
+
+def _load_frontmatter(path: Path) -> tuple[str, dict[str, str]]:
+    """
+    Load content and tags from a file with optional YAML frontmatter.
+
+    Args:
+        path: Path to the file
+
+    Returns:
+        (content, tags) tuple. Tags empty if no frontmatter.
+
+    Raises:
+        FileNotFoundError: If the file doesn't exist
+    """
+    text = path.read_text()
+
+    # Parse YAML frontmatter if present
+    if text.startswith("---"):
+        parts = text.split("---", 2)
+        if len(parts) >= 3:
+            import yaml
+            frontmatter = yaml.safe_load(parts[1])
+            content = parts[2].lstrip("\n")
+            if frontmatter:
+                tags = frontmatter.get("tags", {})
+                # Ensure all tag values are strings
+                tags = {k: str(v) for k, v in tags.items()}
+                return content, tags
+            return content, {}
+
+    return text, {}
+
+
+def _get_env_tags() -> dict[str, str]:
+    """
+    Collect tags from KEEP_TAG_* environment variables.
+
+    KEEP_TAG_PROJECT=foo -> {"project": "foo"}
+    KEEP_TAG_MyTag=bar   -> {"mytag": "bar"}
+
+    Tag keys are lowercased for consistency.
+    """
+    tags = {}
+    for key, value in os.environ.items():
+        if key.startswith(ENV_TAG_PREFIX) and value:
+            tag_key = key[len(ENV_TAG_PREFIX):].lower()
+            tags[tag_key] = value
+    return tags
+
+
+def _content_hash(content: str) -> str:
+    """SHA256 hash of content for change detection."""
+    return hashlib.sha256(content.encode("utf-8")).hexdigest()
+
 
 class Keeper:
     """
@@ -60,20 +210,15 @@ class Keeper:
     ) -> None:
         """
         Initialize or open an existing associative memory store.
-        
+
         Args:
             store_path: Path to store directory. Uses default if not specified.
+                       Overrides any store.path setting in config.
             collection: Default collection name.
             decay_half_life_days: Memory decay half-life in days (ACT-R model).
                 After this many days, an item's effective relevance is halved.
                 Set to 0 or negative to disable decay.
         """
-        # Resolve store path
-        if store_path is None:
-            self._store_path = get_default_store_path()
-        else:
-            self._store_path = Path(store_path).resolve()
-        
         # Validate collection name
         if not COLLECTION_NAME_PATTERN.match(collection):
             raise ValueError(
@@ -82,43 +227,161 @@ class Keeper:
             )
         self._default_collection = collection
         self._decay_half_life_days = decay_half_life_days
-        
+
+        # Resolve config and store paths
+        # If store_path is explicitly provided, use it as both config and store location
+        # Otherwise, discover config via tree-walk and let config determine store
+        if store_path is not None:
+            self._store_path = Path(store_path).resolve()
+            config_dir = self._store_path
+        else:
+            # Discover config directory (tree-walk or envvar)
+            config_dir = get_config_dir()
+
         # Load or create configuration
-        self._config: StoreConfig = load_or_create_config(self._store_path)
-        
-        # Initialize providers
+        self._config: StoreConfig = load_or_create_config(config_dir)
+
+        # If store_path wasn't explicit, resolve from config
+        if store_path is None:
+            self._store_path = get_default_store_path(self._config)
+
+        # Initialize document provider (needed for most operations)
         registry = get_registry()
-        
         self._document_provider: DocumentProvider = registry.create_document(
             self._config.document.name,
             self._config.document.params,
         )
-        
-        # Create embedding provider with caching
-        base_embedding_provider = registry.create_embedding(
-            self._config.embedding.name,
-            self._config.embedding.params,
-        )
-        cache_path = self._store_path / "embedding_cache.db"
-        self._embedding_provider: EmbeddingProvider = CachingEmbeddingProvider(
-            base_embedding_provider,
-            cache_path=cache_path,
-        )
-        
-        self._summarization_provider: SummarizationProvider = registry.create_summarization(
-            self._config.summarization.name,
-            self._config.summarization.params,
-        )
+
+        # Lazy-loaded providers (created on first use to avoid network access for read-only ops)
+        self._embedding_provider: Optional[EmbeddingProvider] = None
+        self._summarization_provider: Optional[SummarizationProvider] = None
 
         # Initialize pending summary queue
         queue_path = self._store_path / "pending_summaries.db"
         self._pending_queue = PendingSummaryQueue(queue_path)
 
-        # Initialize store
+        # Initialize document store (canonical records)
+        from .document_store import DocumentStore
+        doc_store_path = self._store_path / "documents.db"
+        self._document_store = DocumentStore(doc_store_path)
+
+        # Initialize ChromaDB store (embedding index)
+        # Use dimension from stored identity if available (allows offline read-only access)
+        embedding_dim = None
+        if self._config.embedding_identity:
+            embedding_dim = self._config.embedding_identity.dimension
         self._store = ChromaStore(
             self._store_path,
-            embedding_dimension=self._embedding_provider.dimension,
+            embedding_dimension=embedding_dim,
+        )
+
+        # Preload system documents (only if not already present)
+        self._ensure_system_documents()
+
+    def _ensure_system_documents(self) -> None:
+        """
+        Ensure system documents are loaded into the store.
+
+        Scans all .md files in docs/system/. Each file is indexed with its
+        file:// URI as the ID and `_category: system` tag for identification.
+        Content becomes the summary directly (no auto-summarization).
+
+        Called during init. Only loads docs that don't already exist,
+        so user modifications are preserved and no network access occurs
+        if docs are already present.
+        """
+        for path in SYSTEM_DOC_DIR.glob("*.md"):
+            try:
+                uri = f"file://{path.resolve()}"
+                if not self.exists(uri):
+                    content, tags = _load_frontmatter(path)
+                    tags["category"] = "system"
+                    self.remember(content, id=uri, tags=tags)
+            except FileNotFoundError:
+                # System file missing - skip silently
+                pass
+
+    def _get_embedding_provider(self) -> EmbeddingProvider:
+        """
+        Get embedding provider, creating it lazily on first use.
+
+        This allows read-only operations to work offline without loading
+        the embedding model (which may try to reach HuggingFace).
+        """
+        if self._embedding_provider is None:
+            registry = get_registry()
+            base_provider = registry.create_embedding(
+                self._config.embedding.name,
+                self._config.embedding.params,
+            )
+            cache_path = self._store_path / "embedding_cache.db"
+            self._embedding_provider = CachingEmbeddingProvider(
+                base_provider,
+                cache_path=cache_path,
+            )
+            # Validate or record embedding identity
+            self._validate_embedding_identity(self._embedding_provider)
+            # Update store's embedding dimension if it wasn't known at init
+            if self._store._embedding_dimension is None:
+                self._store._embedding_dimension = self._embedding_provider.dimension
+        return self._embedding_provider
+
+    def _get_summarization_provider(self) -> SummarizationProvider:
+        """
+        Get summarization provider, creating it lazily on first use.
+        """
+        if self._summarization_provider is None:
+            registry = get_registry()
+            self._summarization_provider = registry.create_summarization(
+                self._config.summarization.name,
+                self._config.summarization.params,
+            )
+        return self._summarization_provider
+
+    def _validate_embedding_identity(self, provider: EmbeddingProvider) -> None:
+        """
+        Validate embedding provider matches stored identity, or record it.
+
+        On first use, records the embedding identity to config.
+        On subsequent uses, validates that the current provider matches.
+
+        Raises:
+            ValueError: If embedding provider changed incompatibly
+        """
+        # Get current provider's identity
+        current = EmbeddingIdentity(
+            provider=self._config.embedding.name,
+            model=getattr(provider, "model_name", "unknown"),
+            dimension=provider.dimension,
         )
+
+        stored = self._config.embedding_identity
+
+        if stored is None:
+            # First use: record the identity
+            self._config.embedding_identity = current
+            save_config(self._config)
+        else:
+            # Validate compatibility
+            if (stored.provider != current.provider or
+                stored.model != current.model or
+                stored.dimension != current.dimension):
+                raise ValueError(
+                    f"Embedding provider mismatch!\n"
+                    f"  Stored: {stored.provider}/{stored.model} ({stored.dimension}d)\n"
+                    f"  Current: {current.provider}/{current.model} ({current.dimension}d)\n"
+                    f"\n"
+                    f"Changing embedding providers invalidates existing embeddings.\n"
+                    f"Options:\n"
+                    f"  1. Use the original provider\n"
+                    f"  2. Delete .keep/ and re-index\n"
+                    f"  3. (Future) Run migration to re-embed with new provider"
+                )
+
+    @property
+    def embedding_identity(self) -> EmbeddingIdentity | None:
+        """Current embedding identity (provider, model, dimension)."""
+        return self._config.embedding_identity
     
     def _resolve_collection(self, collection: Optional[str]) -> str:
         """Resolve collection name, validating if provided."""
@@ -135,8 +398,10 @@ class Keeper:
     def update(
         self,
         id: str,
-        source_tags: Optional[dict[str, str]] = None,
+        tags: Optional[dict[str, str]] = None,
         *,
+        summary: Optional[str] = None,
+        source_tags: Optional[dict[str, str]] = None,  # Deprecated alias
         collection: Optional[str] = None,
         lazy: bool = False
     ) -> Item:
@@ -146,84 +411,147 @@ class Keeper:
         Fetches the document, generates embeddings and summary, then stores it.
 
         **Update behavior:**
-        - Summary: Always replaced with newly generated summary
-        - Tags: Merged - existing source tags are preserved, new source_tags override
+        - Summary: Replaced with user-provided or newly generated summary
+        - Tags: Merged - existing tags are preserved, new tags override
           on key collision. System tags (prefixed with _) are always managed by
           the system.
 
         Args:
             id: URI of document to fetch and index
-            source_tags: User-provided tags to merge with existing tags
+            tags: User-provided tags to merge with existing tags
+            summary: User-provided summary (skips auto-summarization if given)
+            source_tags: Deprecated alias for 'tags'
             collection: Target collection (uses default if None)
             lazy: If True, use truncated placeholder summary and queue for
                   background processing. Use `process_pending()` to generate
-                  real summaries later.
+                  real summaries later. Ignored if summary is provided.
 
         Returns:
             The stored Item with merged tags and new summary
         """
+        # Handle deprecated source_tags parameter
+        if source_tags is not None:
+            import warnings
+            warnings.warn(
+                "source_tags is deprecated, use 'tags' instead",
+                DeprecationWarning,
+                stacklevel=2
+            )
+            if tags is None:
+                tags = source_tags
+
         coll = self._resolve_collection(collection)
 
-        # Get existing item to preserve tags
+        # Get existing item to preserve tags (check document store first, fall back to ChromaDB)
         existing_tags = {}
-        existing = self._store.get(coll, id)
-        if existing:
-            # Extract existing non-system tags
-            existing_tags = filter_non_system_tags(existing.tags)
+        existing_doc = self._document_store.get(coll, id)
+        if existing_doc:
+            existing_tags = filter_non_system_tags(existing_doc.tags)
+        else:
+            # Fall back to ChromaDB for legacy data
+            existing = self._store.get(coll, id)
+            if existing:
+                existing_tags = filter_non_system_tags(existing.tags)
 
         # Fetch document
         doc = self._document_provider.fetch(id)
 
+        # Compute content hash for change detection
+        new_hash = _content_hash(doc.content)
+
         # Generate embedding
-        embedding = self._embedding_provider.embed(doc.content)
+        embedding = self._get_embedding_provider().embed(doc.content)
 
-        # Generate summary (or queue for later if lazy)
-        if lazy:
-            # Truncated placeholder
-            if len(doc.content) > TRUNCATE_LENGTH:
-                summary = doc.content[:TRUNCATE_LENGTH] + "..."
+        # Determine summary - skip if content unchanged
+        max_len = self._config.max_summary_length
+        content_unchanged = (
+            existing_doc is not None
+            and existing_doc.content_hash == new_hash
+        )
+
+        if content_unchanged and summary is None:
+            # Content unchanged - preserve existing summary
+            logger.debug("Content unchanged, skipping summarization for %s", id)
+            final_summary = existing_doc.summary
+        elif summary is not None:
+            # User-provided summary - validate length
+            if len(summary) > max_len:
+                import warnings
+                warnings.warn(
+                    f"Summary exceeds max_summary_length ({len(summary)} > {max_len}), truncating",
+                    UserWarning,
+                    stacklevel=2
+                )
+                summary = summary[:max_len]
+            final_summary = summary
+        elif lazy:
+            # Truncated placeholder for lazy mode
+            if len(doc.content) > max_len:
+                final_summary = doc.content[:max_len] + "..."
             else:
-                summary = doc.content
+                final_summary = doc.content
             # Queue for background processing
             self._pending_queue.enqueue(id, coll, doc.content)
         else:
-            summary = self._summarization_provider.summarize(doc.content)
+            # Auto-generate summary
+            final_summary = self._get_summarization_provider().summarize(doc.content)
 
-        # Build tags: existing + new (new overrides on collision)
-        tags = {**existing_tags}
+        # Build tags: existing → config → env → user (later wins on collision)
+        merged_tags = {**existing_tags}
 
-        # Merge in new source tags (filtered to prevent system tag override)
-        if source_tags:
-            tags.update(filter_non_system_tags(source_tags))
+        # Merge config default tags
+        if self._config.default_tags:
+            merged_tags.update(self._config.default_tags)
+
+        # Merge environment variable tags
+        env_tags = _get_env_tags()
+        merged_tags.update(env_tags)
+
+        # Merge in user-provided tags (filtered to prevent system tag override)
+        if tags:
+            merged_tags.update(filter_non_system_tags(tags))
 
         # Add system tags
-        tags["_source"] = "uri"
+        merged_tags["_source"] = "uri"
         if doc.content_type:
-            tags["_content_type"] = doc.content_type
+            merged_tags["_content_type"] = doc.content_type
 
-        # Store
+        # Dual-write: document store (canonical) + ChromaDB (embedding index)
+        self._document_store.upsert(
+            collection=coll,
+            id=id,
+            summary=final_summary,
+            tags=merged_tags,
+            content_hash=new_hash,
+        )
         self._store.upsert(
             collection=coll,
             id=id,
             embedding=embedding,
-            summary=summary,
-            tags=tags,
+            summary=final_summary,
+            tags=merged_tags,
         )
 
-        # Spawn background processor if lazy
-        if lazy:
+        # Spawn background processor if lazy (only if summary wasn't user-provided and content changed)
+        if lazy and summary is None and not content_unchanged:
             self._spawn_processor()
 
         # Return the stored item
-        result = self._store.get(coll, id)
-        return result.to_item()
+        doc_record = self._document_store.get(coll, id)
+        return Item(
+            id=doc_record.id,
+            summary=doc_record.summary,
+            tags=doc_record.tags,
+        )
     
     def remember(
         self,
         content: str,
         *,
         id: Optional[str] = None,
-        source_tags: Optional[dict[str, str]] = None,
+        summary: Optional[str] = None,
+        tags: Optional[dict[str, str]] = None,
+        source_tags: Optional[dict[str, str]] = None,  # Deprecated alias
         collection: Optional[str] = None,
         lazy: bool = False
     ) -> Item:
@@ -232,24 +560,42 @@ class Keeper:
 
         Use for conversation snippets, notes, insights.
 
+        **Smart summary behavior:**
+        - If summary is provided, use it (skips auto-summarization)
+        - If content is short (≤ max_summary_length), use content verbatim
+        - Otherwise, generate summary via summarization provider
+
         **Update behavior (when id already exists):**
-        - Summary: Replaced with newly generated summary from content
-        - Tags: Merged - existing source tags preserved, new source_tags override
+        - Summary: Replaced with user-provided, content, or generated summary
+        - Tags: Merged - existing tags preserved, new tags override
           on key collision. System tags (prefixed with _) are always managed by
           the system.
 
         Args:
             content: Text to store and index
             id: Optional custom ID (auto-generated if None)
-            source_tags: User-provided tags to merge with existing tags
+            summary: User-provided summary (skips auto-summarization if given)
+            tags: User-provided tags to merge with existing tags
+            source_tags: Deprecated alias for 'tags'
             collection: Target collection (uses default if None)
-            lazy: If True, use truncated placeholder summary and queue for
-                  background processing. Use `process_pending()` to generate
-                  real summaries later.
+            lazy: If True and content is long, use truncated placeholder summary
+                  and queue for background processing. Ignored if content is
+                  short or summary is provided.
 
         Returns:
             The stored Item with merged tags and new summary
         """
+        # Handle deprecated source_tags parameter
+        if source_tags is not None:
+            import warnings
+            warnings.warn(
+                "source_tags is deprecated, use 'tags' instead",
+                DeprecationWarning,
+                stacklevel=2
+            )
+            if tags is None:
+                tags = source_tags
+
         coll = self._resolve_collection(collection)
 
         # Generate ID if not provided
@@ -257,54 +603,101 @@ class Keeper:
             timestamp = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%S.%f")
             id = f"mem:{timestamp}"
 
-        # Get existing item to preserve tags
+        # Get existing item to preserve tags (check document store first, fall back to ChromaDB)
         existing_tags = {}
-        existing = self._store.get(coll, id)
-        if existing:
-            # Extract existing non-system tags
-            existing_tags = filter_non_system_tags(existing.tags)
+        existing_doc = self._document_store.get(coll, id)
+        if existing_doc:
+            existing_tags = filter_non_system_tags(existing_doc.tags)
+        else:
+            existing = self._store.get(coll, id)
+            if existing:
+                existing_tags = filter_non_system_tags(existing.tags)
+
+        # Compute content hash for change detection
+        new_hash = _content_hash(content)
 
         # Generate embedding
-        embedding = self._embedding_provider.embed(content)
+        embedding = self._get_embedding_provider().embed(content)
 
-        # Generate summary (or queue for later if lazy)
-        if lazy:
-            # Truncated placeholder
-            if len(content) > TRUNCATE_LENGTH:
-                summary = content[:TRUNCATE_LENGTH] + "..."
-            else:
-                summary = content
+        # Determine summary (smart behavior for remember) - skip if content unchanged
+        max_len = self._config.max_summary_length
+        content_unchanged = (
+            existing_doc is not None
+            and existing_doc.content_hash == new_hash
+        )
+
+        if content_unchanged and summary is None:
+            # Content unchanged - preserve existing summary
+            logger.debug("Content unchanged, skipping summarization for %s", id)
+            final_summary = existing_doc.summary
+        elif summary is not None:
+            # User-provided summary - validate length
+            if len(summary) > max_len:
+                import warnings
+                warnings.warn(
+                    f"Summary exceeds max_summary_length ({len(summary)} > {max_len}), truncating",
+                    UserWarning,
+                    stacklevel=2
+                )
+                summary = summary[:max_len]
+            final_summary = summary
+        elif len(content) <= max_len:
+            # Content is short enough - use verbatim (smart summary)
+            final_summary = content
+        elif lazy:
+            # Content is long and lazy mode - truncated placeholder
+            final_summary = content[:max_len] + "..."
             # Queue for background processing
             self._pending_queue.enqueue(id, coll, content)
         else:
-            summary = self._summarization_provider.summarize(content)
+            # Content is long - generate summary
+            final_summary = self._get_summarization_provider().summarize(content)
+
+        # Build tags: existing → config → env → user (later wins on collision)
+        merged_tags = {**existing_tags}
 
-        # Build tags: existing + new (new overrides on collision)
-        tags = {**existing_tags}
+        # Merge config default tags
+        if self._config.default_tags:
+            merged_tags.update(self._config.default_tags)
 
-        # Merge in new source tags (filtered)
-        if source_tags:
-            tags.update(filter_non_system_tags(source_tags))
+        # Merge environment variable tags
+        env_tags = _get_env_tags()
+        merged_tags.update(env_tags)
+
+        # Merge in user-provided tags (filtered)
+        if tags:
+            merged_tags.update(filter_non_system_tags(tags))
 
         # Add system tags
-        tags["_source"] = "inline"
+        merged_tags["_source"] = "inline"
 
-        # Store
+        # Dual-write: document store (canonical) + ChromaDB (embedding index)
+        self._document_store.upsert(
+            collection=coll,
+            id=id,
+            summary=final_summary,
+            tags=merged_tags,
+            content_hash=new_hash,
+        )
         self._store.upsert(
             collection=coll,
             id=id,
             embedding=embedding,
-            summary=summary,
-            tags=tags,
+            summary=final_summary,
+            tags=merged_tags,
         )
 
-        # Spawn background processor if lazy
-        if lazy:
+        # Spawn background processor if lazy and content was queued (only if content changed)
+        if lazy and summary is None and len(content) > max_len and not content_unchanged:
             self._spawn_processor()
 
         # Return the stored item
-        result = self._store.get(coll, id)
-        return result.to_item()
+        doc_record = self._document_store.get(coll, id)
+        return Item(
+            id=doc_record.id,
+            summary=doc_record.summary,
+            tags=doc_record.tags,
+        )
 
     # -------------------------------------------------------------------------
     # Query Operations
@@ -361,27 +754,40 @@ class Keeper:
         query: str,
         *,
         limit: int = 10,
+        since: Optional[str] = None,
         collection: Optional[str] = None
     ) -> list[Item]:
         """
         Find items using semantic similarity search.
-        
+
         Scores are adjusted by recency decay (ACT-R model) - older items
         have reduced effective relevance unless recently accessed.
+
+        Args:
+            query: Search query text
+            limit: Maximum results to return
+            since: Only include items updated since (ISO duration like P3D, or date)
+            collection: Target collection
         """
         coll = self._resolve_collection(collection)
-        
+
         # Embed query
-        embedding = self._embedding_provider.embed(query)
-        
-        # Search (fetch extra to account for re-ranking)
+        embedding = self._get_embedding_provider().embed(query)
+
+        # Search (fetch extra to account for re-ranking and date filtering)
         fetch_limit = limit * 2 if self._decay_half_life_days > 0 else limit
+        if since is not None:
+            fetch_limit = max(fetch_limit, limit * 3)  # Fetch more when filtering
         results = self._store.query_embedding(coll, embedding, limit=fetch_limit)
-        
+
         # Convert to Items and apply decay
         items = [r.to_item() for r in results]
         items = self._apply_recency_decay(items)
-        
+
+        # Apply date filter if specified
+        if since is not None:
+            items = _filter_by_date(items, since)
+
         return items[:limit]
     
     def find_similar(
@@ -389,32 +795,46 @@ class Keeper:
         id: str,
         *,
         limit: int = 10,
+        since: Optional[str] = None,
         include_self: bool = False,
         collection: Optional[str] = None
     ) -> list[Item]:
         """
         Find items similar to an existing item.
+
+        Args:
+            id: ID of item to find similar items for
+            limit: Maximum results to return
+            since: Only include items updated since (ISO duration like P3D, or date)
+            include_self: Include the queried item in results
+            collection: Target collection
         """
         coll = self._resolve_collection(collection)
-        
+
         # Get the item to find its embedding
         item = self._store.get(coll, id)
         if item is None:
             raise KeyError(f"Item not found: {id}")
-        
-        # Search using the summary's embedding
-        embedding = self._embedding_provider.embed(item.summary)
+
+        # Search using the summary's embedding (fetch extra when filtering)
+        embedding = self._get_embedding_provider().embed(item.summary)
         actual_limit = limit + 1 if not include_self else limit
+        if since is not None:
+            actual_limit = max(actual_limit, limit * 3)
         results = self._store.query_embedding(coll, embedding, limit=actual_limit)
-        
+
         # Filter self if needed
         if not include_self:
             results = [r for r in results if r.id != id]
-        
+
         # Convert to Items and apply decay
         items = [r.to_item() for r in results]
         items = self._apply_recency_decay(items)
-        
+
+        # Apply date filter if specified
+        if since is not None:
+            items = _filter_by_date(items, since)
+
         return items[:limit]
     
     def query_fulltext(
@@ -422,14 +842,30 @@ class Keeper:
         query: str,
         *,
         limit: int = 10,
+        since: Optional[str] = None,
         collection: Optional[str] = None
     ) -> list[Item]:
         """
         Search item summaries using full-text search.
+
+        Args:
+            query: Text to search for in summaries
+            limit: Maximum results to return
+            since: Only include items updated since (ISO duration like P3D, or date)
+            collection: Target collection
         """
         coll = self._resolve_collection(collection)
-        results = self._store.query_fulltext(coll, query, limit=limit)
-        return [r.to_item() for r in results]
+
+        # Fetch extra when filtering by date
+        fetch_limit = limit * 3 if since is not None else limit
+        results = self._store.query_fulltext(coll, query, limit=fetch_limit)
+        items = [r.to_item() for r in results]
+
+        # Apply date filter if specified
+        if since is not None:
+            items = _filter_by_date(items, since)
+
+        return items[:limit]
     
     def query_tag(
         self,
@@ -437,6 +873,7 @@ class Keeper:
         value: Optional[str] = None,
         *,
         limit: int = 100,
+        since: Optional[str] = None,
         collection: Optional[str] = None,
         **tags: str
     ) -> list[Item]:
@@ -444,21 +881,39 @@ class Keeper:
         Find items by tag(s).
 
         Usage:
-            # Simple: single key-value pair
+            # Key only: find all docs with this tag key (any value)
+            query_tag("project")
+
+            # Key with value: find docs with specific tag value
             query_tag("project", "myapp")
-            query_tag("tradition", "buddhist")
 
-            # Advanced: multiple tags via kwargs
+            # Multiple tags via kwargs
             query_tag(tradition="buddhist", source="mn22")
+
+        Args:
+            key: Tag key to search for
+            value: Tag value (optional, any value if not provided)
+            limit: Maximum results to return
+            since: Only include items updated since (ISO duration like P3D, or date)
+            collection: Target collection
+            **tags: Additional tag filters as keyword arguments
         """
         coll = self._resolve_collection(collection)
 
+        # Key-only query: find docs that have this tag key (any value)
+        # Uses DocumentStore which supports efficient SQL date filtering
+        if key is not None and value is None and not tags:
+            # Convert since to cutoff date for SQL query
+            since_date = _parse_since(since) if since else None
+            docs = self._document_store.query_by_tag_key(
+                coll, key, limit=limit, since_date=since_date
+            )
+            return [Item(id=d.id, summary=d.summary, tags=d.tags) for d in docs]
+
         # Build tag filter from positional or keyword args
         tag_filter = {}
 
-        if key is not None:
-            if value is None:
-                raise ValueError(f"Value required when querying by key '{key}'")
+        if key is not None and value is not None:
             tag_filter[key] = value
 
         if tags:
@@ -467,11 +922,50 @@ class Keeper:
         if not tag_filter:
             raise ValueError("At least one tag must be specified")
 
-        # Build where clause
-        where = {k: v for k, v in tag_filter.items()}
+        # Build where clause for tag filters only
+        # (ChromaDB $gte doesn't support string dates, so date filtering is done post-query)
+        where_conditions = [{k: v} for k, v in tag_filter.items()]
+
+        # Use $and if multiple conditions, otherwise single condition
+        if len(where_conditions) == 1:
+            where = where_conditions[0]
+        else:
+            where = {"$and": where_conditions}
+
+        # Fetch extra when filtering by date
+        fetch_limit = limit * 3 if since is not None else limit
+        results = self._store.query_metadata(coll, where, limit=fetch_limit)
+        items = [r.to_item() for r in results]
+
+        # Apply date filter if specified (post-filter)
+        if since is not None:
+            items = _filter_by_date(items, since)
+
+        return items[:limit]
+
+    def list_tags(
+        self,
+        key: Optional[str] = None,
+        *,
+        collection: Optional[str] = None,
+    ) -> list[str]:
+        """
+        List distinct tag keys or values.
 
-        results = self._store.query_metadata(coll, where, limit=limit)
-        return [r.to_item() for r in results]
+        Args:
+            key: If provided, list distinct values for this key.
+                 If None, list distinct tag keys.
+            collection: Target collection
+
+        Returns:
+            Sorted list of distinct keys or values
+        """
+        coll = self._resolve_collection(collection)
+
+        if key is None:
+            return self._document_store.list_distinct_tag_keys(coll)
+        else:
+            return self._document_store.list_distinct_tag_values(coll, key)
     
     # -------------------------------------------------------------------------
     # Direct Access
@@ -480,8 +974,21 @@ class Keeper:
     def get(self, id: str, *, collection: Optional[str] = None) -> Optional[Item]:
         """
         Retrieve a specific item by ID.
+        
+        Reads from document store (canonical), falls back to ChromaDB for legacy data.
         """
         coll = self._resolve_collection(collection)
+        
+        # Try document store first (canonical)
+        doc_record = self._document_store.get(coll, id)
+        if doc_record:
+            return Item(
+                id=doc_record.id,
+                summary=doc_record.summary,
+                tags=doc_record.tags,
+            )
+        
+        # Fall back to ChromaDB for legacy data
         result = self._store.get(coll, id)
         if result is None:
             return None
@@ -492,17 +999,148 @@ class Keeper:
         Check if an item exists in the store.
         """
         coll = self._resolve_collection(collection)
-        return self._store.exists(coll, id)
+        # Check document store first, then ChromaDB
+        return self._document_store.exists(coll, id) or self._store.exists(coll, id)
     
     def delete(self, id: str, *, collection: Optional[str] = None) -> bool:
         """
-        Delete an item from the store.
-        
+        Delete an item from both stores.
+
         Returns True if item existed and was deleted.
         """
         coll = self._resolve_collection(collection)
-        return self._store.delete(coll, id)
-    
+        # Delete from both stores
+        doc_deleted = self._document_store.delete(coll, id)
+        chroma_deleted = self._store.delete(coll, id)
+        return doc_deleted or chroma_deleted
+
+    # -------------------------------------------------------------------------
+    # Current Working Context (Now)
+    # -------------------------------------------------------------------------
+
+    def get_now(self) -> Item:
+        """
+        Get the current working context.
+
+        A singleton document representing what you're currently working on.
+        If it doesn't exist, creates one with default content and tags from
+        docs/system/now.md.
+
+        Returns:
+            The current context Item (never None - auto-creates if missing)
+        """
+        item = self.get(NOWDOC_ID)
+        if item is None:
+            # First-time initialization with default content and tags
+            try:
+                default_content, default_tags = _load_frontmatter(SYSTEM_DOC_DIR / "now.md")
+            except FileNotFoundError:
+                # Fallback if system file is missing
+                default_content = "# Now\n\nYour working context."
+                default_tags = {}
+            item = self.set_now(default_content, tags=default_tags)
+        return item
+
+    def set_now(
+        self,
+        content: str,
+        *,
+        tags: Optional[dict[str, str]] = None,
+    ) -> Item:
+        """
+        Set the current working context.
+
+        Updates the singleton context with new content. Uses remember()
+        internally with the fixed NOWDOC_ID.
+
+        Args:
+            content: New content for the current context
+            tags: Optional additional tags to apply
+
+        Returns:
+            The updated context Item
+        """
+        return self.remember(content, id=NOWDOC_ID, tags=tags)
+
+    def list_system_documents(
+        self,
+        *,
+        collection: Optional[str] = None,
+    ) -> list[Item]:
+        """
+        List all system documents.
+
+        System documents are identified by the `category: system` tag.
+        These are preloaded on init and provide foundational content.
+
+        Args:
+            collection: Target collection (default: default collection)
+
+        Returns:
+            List of system document Items
+        """
+        return self.query_tag("category", "system", collection=collection)
+
+    def tag(
+        self,
+        id: str,
+        tags: Optional[dict[str, str]] = None,
+        *,
+        collection: Optional[str] = None,
+    ) -> Optional[Item]:
+        """
+        Update tags on an existing document without re-processing.
+
+        Does NOT re-fetch, re-embed, or re-summarize. Only updates tags.
+
+        Tag behavior:
+        - Provided tags are merged with existing user tags
+        - Empty string value ("") deletes that tag
+        - System tags (_prefixed) cannot be modified via this method
+
+        Args:
+            id: Document identifier
+            tags: Tags to add/update/delete (empty string = delete)
+            collection: Target collection
+
+        Returns:
+            Updated Item if found, None if document doesn't exist
+        """
+        coll = self._resolve_collection(collection)
+
+        # Get existing item (prefer document store, fall back to ChromaDB)
+        existing = self.get(id, collection=collection)
+        if existing is None:
+            return None
+
+        # Start with existing tags, separate system from user
+        current_tags = dict(existing.tags)
+        system_tags = {k: v for k, v in current_tags.items()
+                       if k.startswith(SYSTEM_TAG_PREFIX)}
+        user_tags = {k: v for k, v in current_tags.items()
+                     if not k.startswith(SYSTEM_TAG_PREFIX)}
+
+        # Apply tag changes (filter out system tags from input)
+        if tags:
+            for key, value in tags.items():
+                if key.startswith(SYSTEM_TAG_PREFIX):
+                    continue  # Cannot modify system tags
+                if value == "":
+                    # Empty string = delete
+                    user_tags.pop(key, None)
+                else:
+                    user_tags[key] = value
+
+        # Merge back: user tags + system tags
+        final_tags = {**user_tags, **system_tags}
+
+        # Dual-write to both stores
+        self._document_store.update_tags(coll, id, final_tags)
+        self._store.update_tags(coll, id, final_tags)
+
+        # Return updated item
+        return self.get(id, collection=collection)
+
     # -------------------------------------------------------------------------
     # Collection Management
     # -------------------------------------------------------------------------
@@ -511,13 +1149,21 @@ class Keeper:
         """
         List all collections in the store.
         """
-        return self._store.list_collections()
+        # Merge collections from both stores
+        doc_collections = set(self._document_store.list_collections())
+        chroma_collections = set(self._store.list_collections())
+        return sorted(doc_collections | chroma_collections)
     
     def count(self, *, collection: Optional[str] = None) -> int:
         """
         Count items in a collection.
+        
+        Returns count from document store if available, else ChromaDB.
         """
         coll = self._resolve_collection(collection)
+        doc_count = self._document_store.count(coll)
+        if doc_count > 0:
+            return doc_count
         return self._store.count(coll)
     
     def embedding_cache_stats(self) -> dict:
@@ -525,7 +1171,10 @@ class Keeper:
         Get embedding cache statistics.
 
         Returns dict with: entries, hits, misses, hit_rate, cache_path
+        Returns {"loaded": False} if embedding provider hasn't been loaded yet.
         """
+        if self._embedding_provider is None:
+            return {"loaded": False}
         if isinstance(self._embedding_provider, CachingEmbeddingProvider):
             return self._embedding_provider.stats()
         return {"enabled": False}
@@ -563,9 +1212,10 @@ class Keeper:
 
             try:
                 # Generate real summary
-                summary = self._summarization_provider.summarize(item.content)
+                summary = self._get_summarization_provider().summarize(item.content)
 
-                # Update the stored item's summary
+                # Update summary in both stores
+                self._document_store.update_summary(item.collection, item.id, summary)
                 self._store.update_summary(item.collection, item.id, summary)
 
                 # Remove from queue
@@ -652,8 +1302,9 @@ class Keeper:
             subprocess.Popen(cmd, **kwargs)
             return True
 
-        except Exception:
-            # Spawn failed - not critical, queue will be processed later
+        except Exception as e:
+            # Spawn failed - log for debugging, queue will be processed later
+            logger.warning("Failed to spawn background processor: %s", e)
             return False
 
     def close(self) -> None:
@@ -662,11 +1313,12 @@ class Keeper:
 
         Good practice to call when done, though Python's GC will clean up eventually.
         """
-        # Close embedding cache if it exists
-        if hasattr(self._embedding_provider, '_cache'):
-            cache = self._embedding_provider._cache
-            if hasattr(cache, 'close'):
-                cache.close()
+        # Close embedding cache if it was loaded
+        if self._embedding_provider is not None:
+            if hasattr(self._embedding_provider, '_cache'):
+                cache = self._embedding_provider._cache
+                if hasattr(cache, 'close'):
+                    cache.close()
 
         # Close pending summary queue
         if hasattr(self, '_pending_queue'):