palaia 2.4.dev1__tar.gz → 2.4.dev3__tar.gz

{palaia-2.4.dev1/palaia.egg-info → palaia-2.4.dev3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: palaia
-Version: 2.4.dev1
+Version: 2.4.dev3
 Summary: Local, cloud-free memory for OpenClaw agents.
 Author-email: byte5 GmbH <hello@byte5.de>
 License: MIT
@@ -51,7 +51,7 @@ Dynamic: license-file
 
 # Palaia — The Knowledge System for AI Agent Teams
 
-**Crash-safe. Local-first. Zero-cloud. The memory system that makes your agents smarter over time.**
+**Your agents forget. Palaia doesn't.**
 
 [![CI](https://github.com/byte5ai/palaia/actions/workflows/ci.yml/badge.svg)](https://github.com/byte5ai/palaia/actions/workflows/ci.yml)
 [![PyPI](https://img.shields.io/pypi/v/palaia)](https://pypi.org/project/palaia/)
@@ -61,6 +61,54 @@ Dynamic: license-file
 
 ---
 
+## What Palaia Does
+
+AI agents are stateless by default. Every session starts from scratch — no memory of past decisions, no shared knowledge between agents, no context that survives a restart.
+
+Palaia gives your agents a persistent, searchable knowledge store. They save what they learn. They find it again by meaning, not keyword. They share it across tools and sessions — automatically.
+
+---
+
+## What Palaia Is Not
+
+- Not a chatbot or prompt manager
+- Not a cloud service (everything runs locally)
+- Not a vector database you manage yourself (it manages itself)
+- Not limited to one tool — works with OpenClaw, Claude Desktop, Cursor, and any MCP client
+
+---
+
+## What You Get
+
+| Capability | What it means |
+|------------|---------------|
+| **Agents remember across sessions** | Knowledge survives restarts, tool switches, and team handoffs |
+| **Find anything by meaning** | Hybrid BM25 + vector search across 6 embedding providers |
+| **Zero-config local setup** | SQLite with native SIMD vector search — no separate database process |
+| **Works everywhere via MCP** | One memory store for OpenClaw, Claude Desktop, Cursor, and more |
+| **Multi-agent ready** | Private, team, and public scopes — agents see what they should |
+| **Crash-safe by default** | SQLite WAL mode survives power loss, kills, OOM |
+| **Fast** | Embed server keeps model in RAM — CLI queries ~1.5s, MCP/Plugin <500ms |
+| **Scales when needed** | Swap to PostgreSQL + pgvector for distributed teams, no code changes |
+
+---
+
+## Comparison
+
+| Feature | Palaia | claude-mem | Mem0 | Stock Memory |
+|---------|--------|-----------|------|--------------|
+| Local-first | Yes | Yes | No (cloud) | Yes |
+| Cross-tool (MCP) | Yes (any MCP client) | No (Claude Code only) | No | No |
+| Native Vector Search | sqlite-vec / pgvector | ChromaDB (separate) | Cloud | No |
+| Structured Types | memory/process/task | decisions/bugfixes | No | No |
+| Multi-Agent Scopes | private/team/public | No | Per-user | No |
+| Smart Tiering | HOT/WARM/COLD | No | No | No |
+| Embedding Providers | 6 (configurable) | 1 (fixed) | Cloud | None |
+| Open Source | MIT | AGPL-3.0 | Partial | N/A |
+| Crash-safe (WAL) | Yes | Partial | N/A | No |
+
+---
+
 ## Install
 
 ### Recommended: Tell your agent
@@ -122,40 +170,6 @@ palaia status                                        # Check health
 
 ---
 
-## Why Palaia?
-
-| Feature | Details |
-|---------|---------|
-| **Semantic Search** | Hybrid BM25 + vector embeddings. 6 providers: fastembed, sentence-transformers, Ollama, OpenAI, Gemini, BM25. |
-| **Native Vector Search** | sqlite-vec (SIMD KNN) or pgvector (ANN/HNSW). Not Python cosine — real database-level acceleration. |
-| **MCP Server** | `palaia-mcp` — standalone memory for Claude Desktop, Cursor, any MCP host. No OpenClaw required. |
-| **Multi-Backend** | SQLite (default, zero-config) or PostgreSQL + pgvector for distributed teams. |
-| **Crash-Safe** | SQLite WAL mode — survives power loss, kills, OOM. |
-| **Auto-Capture** | OpenClaw plugin captures significant exchanges automatically. |
-| **Structured Types** | memory, process, task — with status, priority, assignee, due date. |
-| **Multi-Agent** | Shared store, scopes (private/team/public), agent aliases, per-agent injection priorities. |
-| **Smart Tiering** | HOT/WARM/COLD rotation based on decay scores and access patterns. |
-| **Embed Server** | Background process holds model in RAM. CLI queries: ~1.5s (was ~3-5s). MCP/Plugin: <500ms. |
-| **Zero-Cloud** | Everything local. No API keys needed for core functionality. |
-
----
-
-## Comparison
-
-| Feature | Palaia | claude-mem | Mem0 | Stock Memory |
-|---------|--------|-----------|------|--------------|
-| Local-first | Yes | Yes | No (cloud) | Yes |
-| Cross-tool (MCP) | Yes (any MCP client) | No (Claude Code only) | No | No |
-| Native Vector Search | sqlite-vec / pgvector | ChromaDB (separate) | Cloud | No |
-| Structured Types | memory/process/task | decisions/bugfixes | No | No |
-| Multi-Agent Scopes | private/team/public | No | Per-user | No |
-| Smart Tiering | HOT/WARM/COLD | No | No | No |
-| Embedding Providers | 6 (configurable) | 1 (fixed) | Cloud | None |
-| Open Source | MIT | AGPL-3.0 | Partial | N/A |
-| Crash-safe (WAL) | Yes | Partial | N/A | No |
-
----
-
 ## Documentation
 
 | Document | Description |

{palaia-2.4.dev1 → palaia-2.4.dev3}/README.md

@@ -1,6 +1,6 @@
 # Palaia — The Knowledge System for AI Agent Teams
 
-**Crash-safe. Local-first. Zero-cloud. The memory system that makes your agents smarter over time.**
+**Your agents forget. Palaia doesn't.**
 
 [![CI](https://github.com/byte5ai/palaia/actions/workflows/ci.yml/badge.svg)](https://github.com/byte5ai/palaia/actions/workflows/ci.yml)
 [![PyPI](https://img.shields.io/pypi/v/palaia)](https://pypi.org/project/palaia/)
@@ -10,6 +10,54 @@
 
 ---
 
+## What Palaia Does
+
+AI agents are stateless by default. Every session starts from scratch — no memory of past decisions, no shared knowledge between agents, no context that survives a restart.
+
+Palaia gives your agents a persistent, searchable knowledge store. They save what they learn. They find it again by meaning, not keyword. They share it across tools and sessions — automatically.
+
+---
+
+## What Palaia Is Not
+
+- Not a chatbot or prompt manager
+- Not a cloud service (everything runs locally)
+- Not a vector database you manage yourself (it manages itself)
+- Not limited to one tool — works with OpenClaw, Claude Desktop, Cursor, and any MCP client
+
+---
+
+## What You Get
+
+| Capability | What it means |
+|------------|---------------|
+| **Agents remember across sessions** | Knowledge survives restarts, tool switches, and team handoffs |
+| **Find anything by meaning** | Hybrid BM25 + vector search across 6 embedding providers |
+| **Zero-config local setup** | SQLite with native SIMD vector search — no separate database process |
+| **Works everywhere via MCP** | One memory store for OpenClaw, Claude Desktop, Cursor, and more |
+| **Multi-agent ready** | Private, team, and public scopes — agents see what they should |
+| **Crash-safe by default** | SQLite WAL mode survives power loss, kills, OOM |
+| **Fast** | Embed server keeps model in RAM — CLI queries ~1.5s, MCP/Plugin <500ms |
+| **Scales when needed** | Swap to PostgreSQL + pgvector for distributed teams, no code changes |
+
+---
+
+## Comparison
+
+| Feature | Palaia | claude-mem | Mem0 | Stock Memory |
+|---------|--------|-----------|------|--------------|
+| Local-first | Yes | Yes | No (cloud) | Yes |
+| Cross-tool (MCP) | Yes (any MCP client) | No (Claude Code only) | No | No |
+| Native Vector Search | sqlite-vec / pgvector | ChromaDB (separate) | Cloud | No |
+| Structured Types | memory/process/task | decisions/bugfixes | No | No |
+| Multi-Agent Scopes | private/team/public | No | Per-user | No |
+| Smart Tiering | HOT/WARM/COLD | No | No | No |
+| Embedding Providers | 6 (configurable) | 1 (fixed) | Cloud | None |
+| Open Source | MIT | AGPL-3.0 | Partial | N/A |
+| Crash-safe (WAL) | Yes | Partial | N/A | No |
+
+---
+
 ## Install
 
 ### Recommended: Tell your agent
@@ -71,40 +119,6 @@ palaia status                                        # Check health
 
 ---
 
-## Why Palaia?
-
-| Feature | Details |
-|---------|---------|
-| **Semantic Search** | Hybrid BM25 + vector embeddings. 6 providers: fastembed, sentence-transformers, Ollama, OpenAI, Gemini, BM25. |
-| **Native Vector Search** | sqlite-vec (SIMD KNN) or pgvector (ANN/HNSW). Not Python cosine — real database-level acceleration. |
-| **MCP Server** | `palaia-mcp` — standalone memory for Claude Desktop, Cursor, any MCP host. No OpenClaw required. |
-| **Multi-Backend** | SQLite (default, zero-config) or PostgreSQL + pgvector for distributed teams. |
-| **Crash-Safe** | SQLite WAL mode — survives power loss, kills, OOM. |
-| **Auto-Capture** | OpenClaw plugin captures significant exchanges automatically. |
-| **Structured Types** | memory, process, task — with status, priority, assignee, due date. |
-| **Multi-Agent** | Shared store, scopes (private/team/public), agent aliases, per-agent injection priorities. |
-| **Smart Tiering** | HOT/WARM/COLD rotation based on decay scores and access patterns. |
-| **Embed Server** | Background process holds model in RAM. CLI queries: ~1.5s (was ~3-5s). MCP/Plugin: <500ms. |
-| **Zero-Cloud** | Everything local. No API keys needed for core functionality. |
-
----
-
-## Comparison
-
-| Feature | Palaia | claude-mem | Mem0 | Stock Memory |
-|---------|--------|-----------|------|--------------|
-| Local-first | Yes | Yes | No (cloud) | Yes |
-| Cross-tool (MCP) | Yes (any MCP client) | No (Claude Code only) | No | No |
-| Native Vector Search | sqlite-vec / pgvector | ChromaDB (separate) | Cloud | No |
-| Structured Types | memory/process/task | decisions/bugfixes | No | No |
-| Multi-Agent Scopes | private/team/public | No | Per-user | No |
-| Smart Tiering | HOT/WARM/COLD | No | No | No |
-| Embedding Providers | 6 (configurable) | 1 (fixed) | Cloud | None |
-| Open Source | MIT | AGPL-3.0 | Partial | N/A |
-| Crash-safe (WAL) | Yes | Partial | N/A | No |
-
----
-
 ## Documentation
 
 | Document | Description |

{palaia-2.4.dev1 → palaia-2.4.dev3}/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: palaia
-version: "2.4.dev1"
+version: "2.4.dev3"
 description: >
   Local, crash-safe persistent memory for OpenClaw agents.
   SQLite-backed by default. Semantic search, projects, scopes, auto-capture.
@@ -25,17 +25,18 @@ metadata:
       - command: "python3 -m pip install --upgrade 'palaia[fastembed]' && (npm install -g @byte5ai/palaia@latest 2>/dev/null || true) && palaia doctor --fix && palaia warmup"
         label: "Upgrade Palaia with semantic search (pip + npm plugin) and run health checks"
     postUpdateMessage: >
-      Palaia has been updated to v2.2. New: SQLite default backend, injection priorities,
-      knowledge curation, ContextEngine integration. Run `palaia doctor --fix` to upgrade.
+      Palaia has been updated to v2.4. New: session continuity (automatic briefings and
+      summaries), privacy markers, recency boost, progressive disclosure. Run `palaia doctor --fix` to upgrade.
     plugin:
       slot: memory
       package: "@byte5ai/palaia"
 ---
 
-# Palaia v2.2 — Agent Memory Skill
+# Palaia v2.4 — Agent Memory Skill
 
 Palaia is the memory system for OpenClaw agents. It gives you persistent, searchable, crash-safe memory across sessions. You don't need to understand how it works — just use the commands below.
 
+<!-- begin:install -->
 ## First-Time Setup
 
 When a user asks you to "set up palaia", "enable memory", or "remember things across sessions", run these steps:
@@ -136,6 +137,7 @@ npm install -g @byte5ai/palaia@latest
 ```
 
 After the manual update, `palaia upgrade` will be available for future updates.
+<!-- end:install -->
 
 ## Auto-Check (run every time this skill is loaded)
 
@@ -290,6 +292,9 @@ palaia query "what's the rate limit"
 # Filter by type and status
 palaia query "tasks" --type task --status open
 
+# Filter by tags
+palaia query "session-summary" --tags session-summary
+
 # Search within a project
 palaia query "deploy steps" --project myapp
 
@@ -324,6 +329,9 @@ palaia list
 # Filter by tier, type, status
 palaia list --tier warm --type task --status open --priority high
 
+# Limit results
+palaia list --type task --status open --limit 5
+
 # Filter by project or assignee
 palaia list --project myapp --assignee Elliot
 ```
@@ -593,7 +601,7 @@ palaia instance set Claw-Main
 | Review accumulated knowledge | `palaia curate analyze` |
 | Share knowledge | `palaia sync export` or `palaia package export` |
 | Check for messages | `palaia memo inbox` |
-| Start of session | `palaia doctor` then `palaia query "active work"` then `palaia memo inbox` |
+| Start of session | Session briefing is now automatic. Just run `palaia doctor` and check `palaia memo inbox`. |
 
 **Do NOT manually write:** facts, decisions, or preferences that came up in the current conversation. Auto-Capture handles these.
 
@@ -632,6 +640,33 @@ palaia init --capture-level <off|minimal|normal|aggressive>
 
 ---
 
+## Session Continuity (NEW in v2.4)
+
+Session continuity gives agents automatic context restoration across sessions. These features work out of the box with the OpenClaw plugin -- no manual setup needed.
+
+### Session Briefings
+On session start, Palaia automatically injects a briefing with the last session summary and any open tasks. This means agents resume work without needing to manually search for context.
+
+### Session Summaries
+When a session ends or resets, Palaia auto-saves a summary of what happened. These are stored as entries with the `session-summary` tag and can be queried:
+```bash
+palaia query "session-summary" --tags session-summary
+```
+
+### Privacy Markers
+Wrap sensitive content in `<private>...</private>` blocks to exclude it from auto-capture. Private blocks are stripped before any extraction runs.
+
+### Recency Boost
+Fresh memories are ranked higher in recall results. The boost factor is configurable via `recallRecencyBoost` (default `0.3`, set to `0` to disable).
+
+### Progressive Disclosure
+When result sets exceed 100 entries, Palaia uses compact mode to keep context manageable. Use `--limit` to control result size explicitly:
+```bash
+palaia list --type task --status open --limit 5
+```
+
+---
+
 ## Plugin Configuration (OpenClaw)
 
 Set in `openclaw.json` under `plugins.entries.palaia.config`:
@@ -649,6 +684,11 @@ Set in `openclaw.json` under `plugins.entries.palaia.config`:
 | `embeddingServer` | `true` | Keep embedding model loaded for fast queries |
 | `showMemorySources` | `true` | Show memory source footnotes |
 | `showCaptureConfirm` | `true` | Show capture confirmations |
+| `sessionSummary` | `true` | Auto-save session summaries on end/reset |
+| `sessionBriefing` | `true` | Load session context on session start |
+| `sessionBriefingMaxChars` | `1500` | Max chars for session briefing injection |
+| `captureToolObservations` | `true` | Track tool usage as session context |
+| `recallRecencyBoost` | `0.3` | Boost factor for fresh memories (0=off) |
 
 ---
 

{palaia-2.4.dev1 → palaia-2.4.dev3}/palaia/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: palaia
-version: "2.4.dev1"
+version: "2.4.dev3"
 description: >
   Local, crash-safe persistent memory for OpenClaw agents.
   SQLite-backed by default. Semantic search, projects, scopes, auto-capture.
@@ -25,17 +25,18 @@ metadata:
       - command: "python3 -m pip install --upgrade 'palaia[fastembed]' && (npm install -g @byte5ai/palaia@latest 2>/dev/null || true) && palaia doctor --fix && palaia warmup"
         label: "Upgrade Palaia with semantic search (pip + npm plugin) and run health checks"
     postUpdateMessage: >
-      Palaia has been updated to v2.2. New: SQLite default backend, injection priorities,
-      knowledge curation, ContextEngine integration. Run `palaia doctor --fix` to upgrade.
+      Palaia has been updated to v2.4. New: session continuity (automatic briefings and
+      summaries), privacy markers, recency boost, progressive disclosure. Run `palaia doctor --fix` to upgrade.
     plugin:
       slot: memory
       package: "@byte5ai/palaia"
 ---
 
-# Palaia v2.2 — Agent Memory Skill
+# Palaia v2.4 — Agent Memory Skill
 
 Palaia is the memory system for OpenClaw agents. It gives you persistent, searchable, crash-safe memory across sessions. You don't need to understand how it works — just use the commands below.
 
+<!-- begin:install -->
 ## First-Time Setup
 
 When a user asks you to "set up palaia", "enable memory", or "remember things across sessions", run these steps:
@@ -136,6 +137,7 @@ npm install -g @byte5ai/palaia@latest
 ```
 
 After the manual update, `palaia upgrade` will be available for future updates.
+<!-- end:install -->
 
 ## Auto-Check (run every time this skill is loaded)
 
@@ -290,6 +292,9 @@ palaia query "what's the rate limit"
 # Filter by type and status
 palaia query "tasks" --type task --status open
 
+# Filter by tags
+palaia query "session-summary" --tags session-summary
+
 # Search within a project
 palaia query "deploy steps" --project myapp
 
@@ -324,6 +329,9 @@ palaia list
 # Filter by tier, type, status
 palaia list --tier warm --type task --status open --priority high
 
+# Limit results
+palaia list --type task --status open --limit 5
+
 # Filter by project or assignee
 palaia list --project myapp --assignee Elliot
 ```
@@ -593,7 +601,7 @@ palaia instance set Claw-Main
 | Review accumulated knowledge | `palaia curate analyze` |
 | Share knowledge | `palaia sync export` or `palaia package export` |
 | Check for messages | `palaia memo inbox` |
-| Start of session | `palaia doctor` then `palaia query "active work"` then `palaia memo inbox` |
+| Start of session | Session briefing is now automatic. Just run `palaia doctor` and check `palaia memo inbox`. |
 
 **Do NOT manually write:** facts, decisions, or preferences that came up in the current conversation. Auto-Capture handles these.
 
@@ -632,6 +640,33 @@ palaia init --capture-level <off|minimal|normal|aggressive>
 
 ---
 
+## Session Continuity (NEW in v2.4)
+
+Session continuity gives agents automatic context restoration across sessions. These features work out of the box with the OpenClaw plugin -- no manual setup needed.
+
+### Session Briefings
+On session start, Palaia automatically injects a briefing with the last session summary and any open tasks. This means agents resume work without needing to manually search for context.
+
+### Session Summaries
+When a session ends or resets, Palaia auto-saves a summary of what happened. These are stored as entries with the `session-summary` tag and can be queried:
+```bash
+palaia query "session-summary" --tags session-summary
+```
+
+### Privacy Markers
+Wrap sensitive content in `<private>...</private>` blocks to exclude it from auto-capture. Private blocks are stripped before any extraction runs.
+
+### Recency Boost
+Fresh memories are ranked higher in recall results. The boost factor is configurable via `recallRecencyBoost` (default `0.3`, set to `0` to disable).
+
+### Progressive Disclosure
+When result sets exceed 100 entries, Palaia uses compact mode to keep context manageable. Use `--limit` to control result size explicitly:
+```bash
+palaia list --type task --status open --limit 5
+```
+
+---
+
 ## Plugin Configuration (OpenClaw)
 
 Set in `openclaw.json` under `plugins.entries.palaia.config`:
@@ -649,6 +684,11 @@ Set in `openclaw.json` under `plugins.entries.palaia.config`:
 | `embeddingServer` | `true` | Keep embedding model loaded for fast queries |
 | `showMemorySources` | `true` | Show memory source footnotes |
 | `showCaptureConfirm` | `true` | Show capture confirmations |
+| `sessionSummary` | `true` | Auto-save session summaries on end/reset |
+| `sessionBriefing` | `true` | Load session context on session start |
+| `sessionBriefingMaxChars` | `1500` | Max chars for session briefing injection |
+| `captureToolObservations` | `true` | Track tool usage as session context |
+| `recallRecencyBoost` | `0.3` | Boost factor for fresh memories (0=off) |
 
 ---
 

{palaia-2.4.dev1 → palaia-2.4.dev3}/palaia/__init__.py

@@ -4,5 +4,5 @@ Palaia — Local, cloud-free memory for OpenClaw agents.
 
 from __future__ import annotations
 
-__version__ = "2.4.dev1"
+__version__ = "2.4.dev3"
 __author__ = "byte5 GmbH"

{palaia-2.4.dev1 → palaia-2.4.dev3}/palaia/cli_args.py

@@ -98,6 +98,7 @@ def build_parser() -> argparse.ArgumentParser:
     )
     p_query.add_argument("--assignee", default=None, help="Filter by assignee")
     p_query.add_argument("--instance", default=None, help="Filter by session identity")
+    p_query.add_argument("--tags", default=None, help="Filter by tags (comma-separated, AND logic)")
     p_query.add_argument("--before", default=None, help="Only entries created before this ISO timestamp")
     p_query.add_argument("--after", default=None, help="Only entries created after this ISO timestamp")
     p_query.add_argument(
@@ -135,6 +136,7 @@ def build_parser() -> argparse.ArgumentParser:
     p_list = sub.add_parser("list", help="List entries in a tier")
     p_list.add_argument("--tier", default=None, choices=["hot", "warm", "cold"], help="Tier to list (default: hot)")
     p_list.add_argument("--all", action="store_true", help="List across all tiers (hot+warm+cold)")
+    p_list.add_argument("--limit", type=int, default=None, help="Max entries to return")
     p_list.add_argument("--project", default=None, help="Filter by project")
     p_list.add_argument("--tag", default=None, action="append", help="Filter by tag (repeatable, AND logic)")
     p_list.add_argument("--scope", default=None, help="Filter by scope")

{palaia-2.4.dev1 → palaia-2.4.dev3}/palaia/cli_commands.py

@@ -58,6 +58,15 @@ def cmd_query(args):
     has_embeddings = svc_result["has_embeddings"]
     bm25_only = svc_result["bm25_only"]
 
+    # Post-filter by --tags if specified (AND logic, comma-separated)
+    tags_filter = getattr(args, "tags", None)
+    if tags_filter:
+        required_tags = [t.strip() for t in tags_filter.split(",") if t.strip()]
+        results = [
+            r for r in results
+            if all(tag in (r.get("tags") or []) for tag in required_tags)
+        ]
+
     # --- Adaptive Nudging for query (Issue #68) ---
     query_nudge_messages = []
     try:
@@ -234,6 +243,11 @@ def cmd_list(args):
     tier_label = svc_result["tier"]
     entries_with_tier = svc_result["entries_with_tier"]
 
+    # Apply --limit if specified
+    limit = getattr(args, "limit", None)
+    if limit is not None and limit > 0:
+        entries_with_tier = entries_with_tier[:limit]
+
     if json_out(
         {
             "tier": tier_label,

{palaia-2.4.dev1 → palaia-2.4.dev3}/palaia/embed_server.py

@@ -242,6 +242,8 @@ class EmbedServer:
         instance = params.get("instance")
         include_cold = params.get("include_cold", False)
         cross_project = params.get("cross_project", False)
+        before = params.get("before")
+        after = params.get("after")
 
         engine = self._bm25_engine if self._warming_up else self.engine
         results = engine.search(
@@ -255,6 +257,8 @@ class EmbedServer:
             priority=priority,
             assignee=assignee,
             instance=instance,
+            before=before,
+            after=after,
             cross_project=cross_project,
         )
 

{palaia-2.4.dev1 → palaia-2.4.dev3}/palaia/mcp/server.py

@@ -123,7 +123,8 @@ def create_server(root: Path, read_only: bool = False) -> FastMCP:
         header_parts = [
             f"ID: {result['id']}",
             f"Title: {meta.get('title', 'untitled')}",
-            f"Type: {meta.get('scope', 'team')}",
+            f"Type: {meta.get('type', 'memory')}",
+            f"Scope: {meta.get('scope', 'team')}",
             f"Tier: {meta.get('tier', '?')}",
             f"Tags: {', '.join(meta.get('tags', []))}",
             f"Created: {meta.get('created', '?')}",

{palaia-2.4.dev1 → palaia-2.4.dev3}/palaia/search.py

@@ -73,6 +73,7 @@ class SearchEngine:
         self._provider = None
         self._index_cache: list[tuple[str, str, dict]] | None = None
         self._index_dirty = True
+        self._index_cache_key: tuple | None = None  # (include_cold, agent) for cache validity
 
     @property
     def provider(self):
@@ -95,7 +96,8 @@ class SearchEngine:
         Uses a cached index when available. Call invalidate_index() after
         write/edit/gc operations to force a rebuild.
         """
-        if not self._index_dirty and self._index_cache is not None:
+        cache_key = (include_cold, agent)
+        if not self._index_dirty and self._index_cache is not None and self._index_cache_key == cache_key:
             # Re-index BM25 from cache (cheap — no disk reads)
             self.bm25.index([(did, text) for did, text, _meta in self._index_cache])
             return list(self._index_cache)
@@ -112,6 +114,7 @@ class SearchEngine:
             docs_with_meta.append((doc_id, full_text, meta))
         self.bm25.index(docs)
         self._index_cache = docs_with_meta
+        self._index_cache_key = cache_key
         self._index_dirty = False
         return docs_with_meta
 

{palaia-2.4.dev1 → palaia-2.4.dev3}/palaia/services/misc.py

@@ -78,7 +78,17 @@ def unlock_project(root: Path, *, project: str) -> dict:
 
 
 def get_skill_content() -> dict:
-    """Read and return SKILL.md content. Returns dict with 'skill' or 'error'."""
+    """Read and return SKILL.md content.
+
+    Strips the install/update section (between ``<!-- begin:install -->``
+    and ``<!-- end:install -->``) because ``palaia skill`` is only called
+    when palaia is already installed — the install block wastes context window.
+    ClawHub and GitHub still see the full file.
+
+    Returns dict with 'skill' or 'error'.
+    """
+    import re
+
     skill_path = Path(__file__).parent.parent / "SKILL.md"
     if not skill_path.exists():
         return {
@@ -88,6 +98,13 @@ def get_skill_content() -> dict:
             )
         }
     content = skill_path.read_text(encoding="utf-8")
+    # Strip install section — agent already has palaia installed
+    content = re.sub(
+        r"<!-- begin:install -->.*?<!-- end:install -->\n*",
+        "",
+        content,
+        flags=re.DOTALL,
+    )
     return {"skill": content}
 
 

{palaia-2.4.dev1 → palaia-2.4.dev3}/palaia/services/query.py

@@ -80,6 +80,10 @@ def search_entries(
                     params["instance"] = instance
                 if cross_project:
                     params["cross_project"] = True
+                if before:
+                    params["before"] = before
+                if after:
+                    params["after"] = after
                 result = client.query(params, timeout=5.0)
                 chain_cfg = _config.get("embedding_chain", [])
                 bm25_only = not chain_cfg or chain_cfg == ["bm25"]

{palaia-2.4.dev1 → palaia-2.4.dev3}/palaia/services/write.py

@@ -62,7 +62,8 @@ def write_entry(
         instance=instance,
     )
 
-    # Check dedup and tier
+    # Check dedup: Store.write() returns existing ID on hash collision
+    # Compare with a fresh read to detect if this was a dedup or new write
     entry = store.read(entry_id)
     tier = "hot"
     actual_scope = scope or "team"
@@ -74,6 +75,10 @@ def write_entry(
             if (root / t / f"{entry_id}.md").exists():
                 tier = t
                 break
+        # If entry existed before write (non-hot tier or older timestamp),
+        # it was deduplicated. Hot + just-created = new entry.
+        if tier != "hot":
+            deduplicated = True
 
     # --- Adaptive Nudging (Issue #68) ---
     nudge_messages: list[str] = []