**Problem:** Cloud projects that exist only in the database (not in
config.json) couldn't be removed. The error "Project 'name' not found"
occurred because config validation failed before database deletion.
**Root cause:**
- In cloud mode, projects can exist in database without config entries
- ProjectService.remove_project() called config_manager.remove_project()
first, which raised ValueError if project wasn't in config
- This prevented removal of cloud-only projects
**Solution:**
1. Check database first for project existence (source of truth)
2. Validate default project status from both database and config
3. Try to remove from config, but catch ValueError if not found
4. Always remove from database if project exists there
**Additional fix:**
- Project list Default column now shows in local mode (always) and in
cloud mode only if default_project_mode is enabled
- This fixes integration tests that expect to see default marker
Fixes cloud-only project deletion bug.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
Updated bisync_commands.py and core_commands.py to use simplified
configure_rclone_remote() function instead of deprecated
add_tenant_to_rclone_config(). Removed MOUNT_PROFILES import and
replaced dynamic help text with static string.
mount_commands.py still has errors but will be removed entirely in Phase 5.
Related to SPEC-20 Phase 2 cleanup.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
Remove deprecated tenant-specific and mount-related functions:
- Removed add_tenant_to_rclone_config() (replaced by configure_rclone_remote)
- Removed remove_tenant_from_rclone_config()
- Removed all mount-related functions:
- RcloneMountProfile class
- MOUNT_PROFILES dict
- get_default_mount_path()
- build_mount_command()
- is_path_mounted()
- get_rclone_processes()
- kill_rclone_process()
- unmount_path()
- cleanup_orphaned_rclone_processes()
Simplified module to only include:
- configure_rclone_remote() for single remote setup
- Core config management (load, save, backup)
Note: mount_commands.py still depends on removed code and will fail to import.
This will be fully resolved in Phase 5 when mount_commands.py is removed entirely.
Part of SPEC-20 Phase 5 (early cleanup).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
Add new configure_rclone_remote() that uses single remote name:
- Single remote: "basic-memory-cloud" (not tenant-specific)
- Simplifies from per-tenant remotes to one credential set per user
- Maintains backup_rclone_config() for safety
- Add comprehensive functional tests for rclone config
- Test remote configuration, updates, save/load operations
Old add_tenant_to_rclone_config() kept for backward compatibility
but will be removed in Phase 5.
Part of SPEC-20 Simplified Project-Scoped Rclone Sync.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
Add CloudProjectConfig model and cloud_projects dict to BasicMemoryConfig:
- CloudProjectConfig tracks local_path, last_sync, bisync_initialized
- cloud_projects: dict[str, CloudProjectConfig] in config
- Fix datetime serialization with mode='json' in model_dump()
- Add comprehensive tests for cloud_projects functionality
- Backward compatible: old configs without cloud_projects load correctly
This enables project-scoped sync configuration tracked in config file
rather than filesystem discovery, preventing phantom projects.
Part of SPEC-20 Simplified Project-Scoped Rclone Sync.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
- Moved config evaluation from module load time to runtime
- Unified add_project command to handle both cloud and local modes
- Commands (default, sync-config, move) now check cloud_mode at runtime
- Fixes test failures where monkeypatch wasn't applied before command registration
The /proxy/projects/projects POST endpoint returns a ProjectStatusResponse
with fields: message, status, default, old_project, new_project.
Updated CloudProjectCreateResponse schema to match this format instead of
expecting name, path, message fields.
Also updated all related tests to use the correct response format:
- tests/cli/test_cloud_utils.py (3 tests)
- tests/cli/test_bisync_commands.py (1 test)
Fixes the validation error when creating cloud projects via upload command:
"bm cloud upload --project test --create-project specs"
Signed-off-by: Pablo Hernandez <pablo@basicmachines.co>
Signed-off-by: phernandez <paul@basicmachines.co>
Reduces JSON payload size by 50-70% for directory-heavy responses by omitting
null fields from serialization.
Changes:
- Added response_model_exclude_none=True to all directory endpoints:
- GET /directory/tree
- GET /directory/structure
- GET /directory/list
Impact:
- Directory nodes no longer serialize 7 null fields (title, permalink,
entity_id, entity_type, content_type, updated_at, file_path)
- For 50+ directories: eliminates 350+ null fields from response
- Payload reduction: ~2.3kb → ~1kb for typical directory trees
- File nodes still include all metadata when present
Example directory node output:
{
"name": "Tools",
"directory_path": "/Tools",
"type": "directory",
"children": []
}
Testing:
- All 29 directory tests passing
- Type checking passing (0 errors)
- Backward compatible (clients just see missing keys vs null)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>
The lifespan-based instrumentation only runs when FastAPI app starts.
In MCP context, the app never starts but the httpx client is still used.
Solution: Instrument the client immediately after creation at module level.
This works in both contexts:
- MCP: client is instrumented when module is imported
- API: client is instrumented before lifespan runs (lifespan still safe)
This enables distributed tracing from MCP -> Cloud -> API.
Signed-off-by: phernandez <paul@basicmachines.co>
phernandez