Add 'mempalace kg' with four sub-actions:
  - kg add <subject> <predicate> <obj> [--source] [--from] [--confidence]
  - kg query <entity> [--as-of] [--direction]
  - kg timeline [<entity>]
  - kg stats

Routes via two-level dispatch (same pattern as 'hook' and 'instructions').
Also restore accidentally dropped 'def cmd_repair(args):' function header.

Co-authored-by: Copilot <[email protected]>

Allows pointing kg subcommands at a specific sqlite3 file,
enabling the KG to live in a project repo and be tracked in git.

Usage: mempalace kg --kg ./mempalace/knowledge_graph.sqlite3 add ...

Co-authored-by: Copilot <[email protected]>

The checkmark character (U+2713) in the progress line caused
UnicodeEncodeError on Windows terminals using cp1252 encoding.

Co-authored-by: Copilot <[email protected]>

Routes .py files through chunk_python_ast() which emits one chunk per
top-level function, class, and method. Each chunk carries symbol_type,
symbol_name, and parent_symbol metadata for filtered retrieval.
Falls back to chunk_text() on SyntaxError or empty AST.

feat(search): min_similarity threshold parameter

search_memories() and the MCP tool_search() accept min_similarity (default
0.0, backward-compatible). Results below threshold are filtered post-query.

Co-authored-by: Copilot <[email protected]>

Adds cursor_miner.py to extract user/assistant exchange pairs from
Cursor's store.db SQLite files (~/.cursor/chats/**/<session>/store.db).

Key design:
- Extracts <user_query> tags from user messages; skips system-context blobs
- Caps assistant text at 1500 chars per exchange for focused retrieval
- Uses rowid order for message sequence (SQLite insertion order)
- Copies DB to temp file before reading to avoid Cursor file locks
- Stores session_name and workspace_hash as metadata for filtering
- Dedup by source_key = 'cursor:<workspace_hash>/<session_hash>'

CLI: mempalace mine ~/.cursor/chats --mode cursor [--wing my_wing]

Co-authored-by: Copilot <[email protected]>

Adds two equivalent scripts (PowerShell + bash) that mine all three AI
chat sources in a single command.

  cursor   → ~/.cursor/chats          --mode cursor  --wing cursor_chats
  copilot  → ~/.copilot/session-state --mode convos  --wing copilot_sessions
  factory  → ~/.factory/sessions      --mode convos  --wing factory_sessions

Both scripts support --dry-run and selective source args.

Co-authored-by: Copilot <[email protected]>

- knowledge_graph: replace shared SQLite connection with thread-local
  storage (threading.local) to fix data-corruption risk in concurrent
  MCP server contexts; remove check_same_thread=False
- mcp_server: hash full content (not content[:100]) for drawer_id to
  eliminate silent ID collisions between entries sharing a common prefix
- hooks/mempal_save_hook.sh: guard TRANSCRIPT_PATH with -n before -f
  to prevent confusing errors when sanitization yields an empty path

Co-authored-by: Copilot <[email protected]>

Includes new palace.py module, expanded mcp_server tools, knowledge
graph refactor, extensive new test suite, CI Windows job, and
integrations/openclaw SKILL.md.

Fixes applied before merge:
- Thread-safe SQLite connections in KnowledgeGraph (threading.local)
- Full-content hashing for drawer IDs in mcp_server
- TRANSCRIPT_PATH empty-string guard in mempal_save_hook.sh

Conflict resolution:
- tests/test_miner.py: merged chunk_python_ast tests (main) with
  file_already_mined tests (PR)
- tests/test_normalize.py: merged copilot/factory normalizer tests
  (main) with PR's extended normalize test suite; combined imports
- tests/test_searcher.py: merged min_similarity tests (main) with
  PR's query-error and filter tests

Co-authored-by: Copilot <[email protected]>

Expand transcript ingestion with Cursor chat mining and broader normalizers so more local AI sessions can be filed into the palace.
Harden Chroma collection access and search ranking so builds and tests pass reliably without network model downloads.

Made-with: Cursor

Mem palace/main

- palace.py: remove duplicate hashlib/re imports; restore get_client,
  get_embedding_function, distance_to_similarity, _SafeEmbeddingFunction,
  and _HashEmbeddingFunction removed by backend refactor
- miner.py: fix indentation — add_drawer() call and if added: block were
  dedented out of the for chunk in chunks: loop
- mcp_server.py: add missing import of get_embedding_function and
  distance_to_similarity from palace
- searcher.py: remove duplicate mid-file import of removed palace symbols
- knowledge_graph.py: add missing self._local = threading.local() init
- tests/conftest.py: remove unused get_client/get_collection imports;
  add import chromadb for fixture at line 103
- tests/test_miner.py: fix syntax error in multi-name import; replace
  get_client with import chromadb (used directly in tests)
- tests/test_convo_miner.py: remove duplicate shutil import; add
  get_collection as get_palace_collection import
- tests/test_mcp_server.py: remove unused get_collection import

Co-authored-by: Copilot <[email protected]>

Mem palace main

…main

# Conflicts:
#	hooks/mempal_save_hook.sh
#	mempalace/cli.py
#	mempalace/layers.py
#	mempalace/miner.py
#	tests/test_hooks_cli.py
#	uv.lock
#	website/.vitepress/config.mts
#	website/.vitepress/theme/landing/HeroSection.vue
#	website/.vitepress/theme/landing/landing.css