`write_note(overwrite=True)` caught the 409 from create_entity and then
called `knowledge_client.resolve_entity(entity.permalink)` with the
default `strict=False`. In workspace-prefixed palaces the client-built
permalink omits the workspace slug, so exact permalink lookup misses
and the fuzzy fallback could pick an orphan row that shares tokens with
the canonical permalink. The update then wrote to the orphan, leaving
the canonical row stale. On the next overwrite the permalink uniqueness
check in `_resolve_schema_permalink` found duplicate rows and minted
`-1`/`-2` suffixes on the canonical entity, accumulating orphans on
every re-synthesis run.
The 409 came from a `file_service.exists(file_path)` check in
`prepare_create_entity_content`, so the file_path is the authoritative
key for the canonical row — no fuzzy matching needed. Resolve by
file_path with `strict=True`, POSIX-normalized so Windows clients send
the form the server stores.
Adds a regression test that spies on `resolve_entity` and asserts the
identifier and `strict` flag, plus checks that no `-1`/`-2` suffix is
minted under the canonical permalink.
Reported in basic-memory-bug-report Issue 1.
Signed-off-by: Dennis Hempel <uhg.dennnis@gmail.com>
The previous backfill trigger relied on Alembic revision tracking, but
alembic_version only stores the head revision — intermediate revisions
(like the backfill trigger) are invisible after a multi-step upgrade or
fresh DB creation.
Three changes fix this:
1. Replace Alembic revision check with a simple "entities exist but
embeddings are empty" check that works regardless of migration path
2. Generate embeddings during sync — after FTS indexing, batch-embed all
synced entities at the end of the sync operation
3. Add background backfill at MCP startup for the upgrade path (entities
already exist, no embeddings) without blocking server readiness
Also adds clear startup logging for semantic embedding status so issues
are easy to spot in the logs.
📋 Covers: fresh DB, upgrade from pre-embedding version, db reset,
interrupted backfill
Signed-off-by: Pedro Hernandez <pedro@basicmachines.co>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
- Add Annotated descriptions to note_types and entity_types parameters so
LLMs can distinguish frontmatter type filtering from knowledge graph item
type filtering (search.py, ui_sdk.py)
- Lowercase note_types values at filter time so "Chapter" matches stored
"chapter"
- Fix misleading entity_types references in schema.py guidance strings
(should be note_types)
- Add permalink pattern documentation note about full path matching
- Add test for note_types case-insensitive lowercasing
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
In Postgres test mode, stale dependency_overrides on the module-level
FastAPI app allow _resolve_default_project_from_api() to query a live
database and return 'test-project' even when the test sets
default_project=None. Monkeypatch the async fallback in the three
affected tests to isolate config-based resolution from API leakage.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
In cloud mode, ConfigManager has no local config so default_project
is always None. Add API fallback in resolve_project_parameter that
queries /v2/projects/ for the default_project field. This fixes all
MCP tools that rely on project resolution (recent_activity, etc).
Removed discovery mode tests that simulated an invalid state by
clearing is_default — there must always be a default project.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
🔧#640 — LinkResolver selects worst match instead of best
Replace `min(results, key=lambda x: x.score)` with `results[0]`.
Both SQLite and Postgres return results sorted best-first in SQL,
so using `results[0]` is backend-agnostic and correct.
🔧#641 — search_notes output_format="text" returns raw Pydantic model
Add `_format_search_markdown()` that formats SearchResponse as readable
markdown with title, permalink, score, and matched snippet per result.
Update prompts to use `output_format="json"` since they need structured
data for result counting and branching logic.
🔧#642 — metadata_filters with `note_type` key returns empty results
Add `_METADATA_KEY_ALIASES` mapping at the tool level that aliases
`note_type` → `type` before passing metadata_filters to the search query.
The frontmatter field is `type`, not `note_type`.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
Update both the compact and extended AI assistant guides with v0.19.0 changes:
- 📝 write_note overwrite guard: callout, edit_note examples, overwrite=True
- 🔍 Expanded search section: all search types, tag: shorthand, filter-only
searches, metadata_filters operators, min_similarity
- ⚠️ "Note already exists" error handling pattern
- 📋 Tool quick reference: updated params, added list_workspaces
- 🔗 memory:// URL: added cross-project format
- ✏️ Best practice: prefer edit_note for updates
Fix tag: shorthand parsing to handle multiple tags anywhere in the query.
Old parser only handled queries starting with "tag:" and broke on
"tag:coffee AND tag:brewing". New parser uses re.findall to extract all
tag:value tokens, strips boolean connectors, and preserves remaining text.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
Add set_workspace_provider() injection point so the cloud MCP server can
list workspaces by querying its own database directly, instead of making
an HTTP round-trip to the control-plane API with credentials it doesn't have.
🔧 Mirrors the existing set_client_factory() pattern in async_client.py
🧪 Adds 3 tests for provider injection, fallback, and context caching
🩹 Updates build_context test assertions for v0.18 backward compat fields
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
Add #634 (stale schema metadata) to bug fixes section.
Apply ruff formatting to schema.py, write_note.py, test files.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
schema_infer and schema_diff returned raw Pydantic models in text mode,
causing LLMs to render field names as "undefined". Add text formatters
(_format_inference_report, _format_drift_report) matching the existing
_format_validation_report pattern. CLI paths are unaffected — they
always use output_format="json".
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
When semantic search is enabled (default), `search_notes(query="tag:security")`
failed because the HYBRID retrieval mode requires non-empty text, but the
service-layer tag: parser clears the text after the mode is already set.
Parse tag: prefix at the tool level before search mode selection, converting it
to a tags filter. This works with all search modes (text, hybrid, vector).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
search_notes was returning individual observations and relations as
separate top-level results, wasting the result limit and creating
confusing UX. Default entity_types to ["entity"] when the caller
doesn't specify it — the entity row already indexes full file content,
so no matches are lost. Users can still override with explicit
entity_types=["observation"] etc.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
Dennis Hempel