fix(cli): defer FastAPI and app imports out of CLI startup

Every basic-memory CLI invocation paid roughly 2 seconds of module-import
cost before any work started, which blew the Claude Code plugin's
SessionStart hook budget on cold machines (#886). The cost came from
module-level imports that pulled the entire server stack into CLI startup:

- mcp/async_client.py imported FastAPI at module level, so every consumer
  of get_client() loaded FastAPI even for cloud-routed or help-only paths.
- mcp/clients/*.py imported call_* helpers from basic_memory.mcp.tools.utils,
  which executes the whole tools package __init__ — every MCP tool module
  plus fastmcp and the mcp SDK.
- mcp/project_context.py imported fastmcp.Context and ToolError eagerly.
- CLI command modules (tool, ci, schema) imported MCP tool functions at
  module level; db and the import_* commands pulled SQLAlchemy/Alembic and
  the markdown/file-service stack; status/doctor/orphans/command_utils
  imported ToolError (the mcp SDK) and basic_memory.db.
- schemas/base.py imported dateparser (~0.13s) for one helper function.

The fix only defers imports to the point of use (no behavior changes):
FastAPI now loads inside _resolve_local_asgi_database alongside the
existing lazy api.app import, so it is only paid when a request actually
routes through the in-process ASGI transport; the typed clients import
call_* per method; project_context uses PEP 563 annotations with Context
under TYPE_CHECKING; the CLI command modules import their heavy
dependencies inside the command bodies. Tests that patched the old
module-level aliases now patch the source modules instead.

Measured on a warm cache (python -X importtime / wall time):
- import basic_memory.cli.main: 1.92s -> 0.45s
- bm --help: 2.40s -> 0.52s
- bm tool search-notes --help: 2.40s -> 0.86s

A regression test asserts that importing the CLI entry module with full
command registration leaves fastapi, sqlalchemy, alembic, fastmcp, mcp,
basic_memory.api.app, basic_memory.db, basic_memory.markdown,
basic_memory.mcp.tools, and basic_memory.services out of sys.modules.

Fixes #886

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Signed-off-by: phernandez <paul@basicmachines.co>

Author: phernandez

src/basic_memory/cli/commands/ci.py

@@ -34,8 +34,10 @@
 from basic_memory.cli.app import app
 from basic_memory.cli.commands.command_utils import run_with_cleanup
 from basic_memory.cli.commands.routing import force_routing, validate_routing_flags
-from basic_memory.mcp.tools import search_notes as mcp_search_notes
-from basic_memory.mcp.tools import write_note as mcp_write_note
+
+# MCP tool functions are imported inside the async helpers below: importing
+# basic_memory.mcp.tools loads the entire tool stack (fastmcp, mcp SDK,
+# SQLAlchemy), which would slow every CLI invocation, including --help (#886).
 
 
 console = Console()
@@ -252,6 +254,10 @@ async def seed_project_update_schemas(
     refresh: bool = False,
 ) -> list[str]:
     """Seed Auto BM schema notes without overwriting customized schemas."""
+    # Deferred: loading the MCP tool stack at module import slows CLI startup (#886).
+    from basic_memory.mcp.tools import search_notes as mcp_search_notes
+    from basic_memory.mcp.tools import write_note as mcp_write_note
+
     seeded: list[str] = []
     routed_project = _routed_project(project=project, project_id=project_id, workspace=workspace)
     for spec in schema_seed_specs():
@@ -295,6 +301,10 @@ async def publish_project_update_note(
     note: ProjectUpdateNote,
 ) -> dict[str, Any]:
     """Search by idempotency key and then upsert the deterministic note path."""
+    # Deferred: loading the MCP tool stack at module import slows CLI startup (#886).
+    from basic_memory.mcp.tools import search_notes as mcp_search_notes
+    from basic_memory.mcp.tools import write_note as mcp_write_note
+
     routed_project = _routed_project(
         project=config.project,
         project_id=config.project_id,

src/basic_memory/cli/commands/command_utils.py

@@ -3,12 +3,10 @@
 import asyncio
 from typing import Optional, TypeVar, Coroutine, Any
 
-from mcp.server.fastmcp.exceptions import ToolError
 import typer
 
 from rich.console import Console
 
-from basic_memory import db
 from basic_memory.config import ConfigManager
 from basic_memory.mcp.async_client import get_client
 from basic_memory.mcp.clients import ProjectClient
@@ -31,6 +29,9 @@ def run_with_cleanup(coro: Coroutine[Any, Any, T]) -> T:
     Returns:
         The result of the coroutine
     """
+    # Deferred: basic_memory.db pulls SQLAlchemy + Alembic, which must not load
+    # at CLI import time — only when a command actually runs (#886).
+    from basic_memory import db
 
     async def _with_cleanup() -> T:
         try:
@@ -53,6 +54,8 @@ async def run_sync(
         force_full: If True, force a full scan bypassing watermark optimization
         run_in_background: If True, return immediately; if False, wait for completion
     """
+    # Deferred: ToolError lives in the mcp SDK, which must not load at CLI startup (#886).
+    from mcp.server.fastmcp.exceptions import ToolError
 
     # Resolve default project so get_client() can route per-project
     project = project or ConfigManager().default_project
@@ -86,6 +89,9 @@ async def run_sync(
 
 async def get_project_info(project: str):
     """Get project information via API endpoint."""
+    # Deferred: ToolError lives in the mcp SDK, which must not load at CLI startup (#886).
+    from mcp.server.fastmcp.exceptions import ToolError
+
     try:
         async with get_client(project_name=project) as client:
             project_item = await get_active_project(client, project, None)

src/basic_memory/cli/commands/db.py

@@ -1,24 +1,27 @@
 """Database management commands."""
 
+# PEP 563 lazy annotations let signatures reference IndexProgress without importing
+# the indexing stack at module load; reset/reindex import their heavy database and
+# sync dependencies at call time so CLI startup stays fast (#886).
+from __future__ import annotations
+
 import os
 from dataclasses import dataclass
 from pathlib import Path, PurePosixPath, PureWindowsPath
+from typing import TYPE_CHECKING
 
 import psutil
 import typer
 from loguru import logger
 from rich.console import Console
 from rich.progress import Progress, SpinnerColumn, TextColumn, BarColumn, TaskProgressColumn
-from sqlalchemy.exc import OperationalError
 
-from basic_memory import db
 from basic_memory.cli.app import app
 from basic_memory.cli.commands.command_utils import run_with_cleanup
 from basic_memory.config import ConfigManager, ProjectMode
-from basic_memory.indexing import IndexProgress
-from basic_memory.repository import ProjectRepository
-from basic_memory.services.initialization import reconcile_projects_with_config
-from basic_memory.sync.sync_service import get_sync_service
+
+if TYPE_CHECKING:
+    from basic_memory.indexing import IndexProgress
 
 console = Console()
 
@@ -159,6 +162,13 @@ async def _reindex_projects(app_config):
     This ensures all database operations use the same event loop,
     and proper cleanup happens when the function completes.
     """
+    # Deferred: SQLAlchemy, repositories, and the sync stack load only when a
+    # reindex actually runs, not on every CLI start (#886).
+    from basic_memory import db
+    from basic_memory.repository import ProjectRepository
+    from basic_memory.services.initialization import reconcile_projects_with_config
+    from basic_memory.sync.sync_service import get_sync_service
+
     try:
         await reconcile_projects_with_config(app_config)
 
@@ -197,6 +207,12 @@ def reset(
     ),
 ):  # pragma: no cover
     """Reset database (drop all tables and recreate)."""
+    # Deferred: SQLAlchemy and the db module load only when a reset actually
+    # runs, not on every CLI start (#886).
+    from sqlalchemy.exc import OperationalError
+
+    from basic_memory import db
+
     console.print(
         "[yellow]Note:[/yellow] This only deletes the index database. "
         "Your markdown note files will not be affected.\n"
@@ -320,10 +336,15 @@ async def _reindex(
     project: str | None,
 ):
     """Run reindex operations."""
-    from basic_memory.repository import EntityRepository
+    # Deferred: SQLAlchemy, repositories, and the sync stack load only when a
+    # reindex actually runs, not on every CLI start (#886).
+    from basic_memory import db
+    from basic_memory.repository import EntityRepository, ProjectRepository
     from basic_memory.repository.search_repository import create_search_repository
+    from basic_memory.services.initialization import reconcile_projects_with_config
     from basic_memory.services.search_service import SearchService
     from basic_memory.services.file_service import FileService
+    from basic_memory.sync.sync_service import get_sync_service
     from basic_memory.markdown.markdown_processor import MarkdownProcessor
     from basic_memory.markdown.entity_parser import EntityParser
 

src/basic_memory/cli/commands/doctor.py

@@ -7,16 +7,12 @@
 from pathlib import Path
 
 from loguru import logger
-from mcp.server.fastmcp.exceptions import ToolError
 from rich.console import Console
 import typer
 
 from basic_memory.cli.app import app
 from basic_memory.cli.commands.command_utils import run_with_cleanup
 from basic_memory.cli.commands.routing import force_routing, validate_routing_flags
-from basic_memory.markdown.entity_parser import EntityParser
-from basic_memory.markdown.markdown_processor import MarkdownProcessor
-from basic_memory.markdown.schemas import EntityFrontmatter, EntityMarkdown
 from basic_memory.mcp.async_client import get_client
 from basic_memory.mcp.clients import KnowledgeClient, ProjectClient, SearchClient
 from basic_memory.schemas.base import Entity
@@ -29,6 +25,12 @@
 
 async def run_doctor() -> None:
     """Run local consistency checks for file <-> database flows."""
+    # Deferred: the markdown parsing stack is only needed while the checks run,
+    # and importing it at module level slows every CLI invocation (#886).
+    from basic_memory.markdown.entity_parser import EntityParser
+    from basic_memory.markdown.markdown_processor import MarkdownProcessor
+    from basic_memory.markdown.schemas import EntityFrontmatter, EntityMarkdown
+
     console.print("[blue]Running Basic Memory doctor checks...[/blue]")
 
     project_name = f"doctor-{uuid.uuid4().hex[:8]}"
@@ -140,6 +142,9 @@ def doctor(
     cloud: bool = typer.Option(False, "--cloud", help="Force cloud API routing"),
 ) -> None:
     """Run local consistency checks to verify file/database sync."""
+    # Deferred: ToolError lives in the mcp SDK, which must not load at CLI startup (#886).
+    from mcp.server.fastmcp.exceptions import ToolError
+
     try:
         validate_routing_flags(local, cloud)
         # Doctor runs local filesystem checks — always default to local routing

src/basic_memory/cli/commands/import_chatgpt.py

@@ -1,25 +1,34 @@
 """Import command for ChatGPT conversations."""
 
+# PEP 563 lazy annotations keep heavy importer types out of module import (#886).
+from __future__ import annotations
+
 import json
 from pathlib import Path
-from typing import Annotated, Tuple
+from typing import TYPE_CHECKING, Annotated, Tuple
 
 import typer
 from basic_memory.cli.app import import_app
 from basic_memory.cli.commands.command_utils import run_with_cleanup
 from basic_memory.config import ConfigManager, get_project_config
-from basic_memory.importers import ChatGPTImporter
-from basic_memory.markdown import EntityParser, MarkdownProcessor
-from basic_memory.services.file_service import FileService
 from loguru import logger
 from rich.console import Console
 from rich.panel import Panel
 
+if TYPE_CHECKING:
+    from basic_memory.markdown import MarkdownProcessor
+    from basic_memory.services.file_service import FileService
+
 console = Console()
 
 
 async def get_importer_dependencies() -> Tuple[MarkdownProcessor, FileService]:
     """Get MarkdownProcessor and FileService instances for importers."""
+    # Deferred: the markdown/file-service stack pulls SQLAlchemy and must load
+    # only when an import actually runs, not on every CLI start (#886).
+    from basic_memory.markdown import EntityParser, MarkdownProcessor
+    from basic_memory.services.file_service import FileService
+
     config = get_project_config()
     app_config = ConfigManager().config
     entity_parser = EntityParser(config.home)
@@ -60,6 +69,9 @@ def import_chatgpt(
         console.print(f"\nImporting chats from {conversations_json}...writing to {base_path}")
 
         # Create importer and run import
+        # Deferred: importer stack loads at import-command run time only (#886).
+        from basic_memory.importers import ChatGPTImporter
+
         importer = ChatGPTImporter(
             config.home, markdown_processor, file_service, project_name=config.name
         )

src/basic_memory/cli/commands/import_claude_conversations.py

@@ -1,25 +1,34 @@
 """Import command for basic-memory CLI to import chat data from conversations2.json format."""
 
+# PEP 563 lazy annotations keep heavy importer types out of module import (#886).
+from __future__ import annotations
+
 import json
 from pathlib import Path
-from typing import Annotated, Tuple
+from typing import TYPE_CHECKING, Annotated, Tuple
 
 import typer
 from basic_memory.cli.app import claude_app
 from basic_memory.cli.commands.command_utils import run_with_cleanup
 from basic_memory.config import ConfigManager, get_project_config
-from basic_memory.importers.claude_conversations_importer import ClaudeConversationsImporter
-from basic_memory.markdown import EntityParser, MarkdownProcessor
-from basic_memory.services.file_service import FileService
 from loguru import logger
 from rich.console import Console
 from rich.panel import Panel
 
+if TYPE_CHECKING:
+    from basic_memory.markdown import MarkdownProcessor
+    from basic_memory.services.file_service import FileService
+
 console = Console()
 
 
 async def get_importer_dependencies() -> Tuple[MarkdownProcessor, FileService]:
     """Get MarkdownProcessor and FileService instances for importers."""
+    # Deferred: the markdown/file-service stack pulls SQLAlchemy and must load
+    # only when an import actually runs, not on every CLI start (#886).
+    from basic_memory.markdown import EntityParser, MarkdownProcessor
+    from basic_memory.services.file_service import FileService
+
     config = get_project_config()
     app_config = ConfigManager().config
     entity_parser = EntityParser(config.home)
@@ -57,6 +66,9 @@ def import_claude(
         markdown_processor, file_service = run_with_cleanup(get_importer_dependencies())
 
         # Create the importer
+        # Deferred: importer stack loads at import-command run time only (#886).
+        from basic_memory.importers.claude_conversations_importer import ClaudeConversationsImporter
+
         importer = ClaudeConversationsImporter(
             config.home, markdown_processor, file_service, project_name=config.name
         )

src/basic_memory/cli/commands/import_claude_projects.py

@@ -1,25 +1,34 @@
 """Import command for basic-memory CLI to import project data from Claude.ai."""
 
+# PEP 563 lazy annotations keep heavy importer types out of module import (#886).
+from __future__ import annotations
+
 import json
 from pathlib import Path
-from typing import Annotated, Tuple
+from typing import TYPE_CHECKING, Annotated, Tuple
 
 import typer
 from basic_memory.cli.app import claude_app
 from basic_memory.cli.commands.command_utils import run_with_cleanup
 from basic_memory.config import ConfigManager, get_project_config
-from basic_memory.importers.claude_projects_importer import ClaudeProjectsImporter
-from basic_memory.markdown import EntityParser, MarkdownProcessor
-from basic_memory.services.file_service import FileService
 from loguru import logger
 from rich.console import Console
 from rich.panel import Panel
 
+if TYPE_CHECKING:
+    from basic_memory.markdown import MarkdownProcessor
+    from basic_memory.services.file_service import FileService
+
 console = Console()
 
 
 async def get_importer_dependencies() -> Tuple[MarkdownProcessor, FileService]:
     """Get MarkdownProcessor and FileService instances for importers."""
+    # Deferred: the markdown/file-service stack pulls SQLAlchemy and must load
+    # only when an import actually runs, not on every CLI start (#886).
+    from basic_memory.markdown import EntityParser, MarkdownProcessor
+    from basic_memory.services.file_service import FileService
+
     config = get_project_config()
     app_config = ConfigManager().config
     entity_parser = EntityParser(config.home)
@@ -56,6 +65,9 @@ def import_projects(
         markdown_processor, file_service = run_with_cleanup(get_importer_dependencies())
 
         # Create the importer
+        # Deferred: importer stack loads at import-command run time only (#886).
+        from basic_memory.importers.claude_projects_importer import ClaudeProjectsImporter
+
         importer = ClaudeProjectsImporter(
             config.home, markdown_processor, file_service, project_name=config.name
         )

src/basic_memory/cli/commands/import_memory_json.py

@@ -1,25 +1,34 @@
 """Import command for basic-memory CLI to import from JSON memory format."""
 
+# PEP 563 lazy annotations keep heavy importer types out of module import (#886).
+from __future__ import annotations
+
 import json
 from pathlib import Path
-from typing import Annotated, Tuple
+from typing import TYPE_CHECKING, Annotated, Tuple
 
 import typer
 from basic_memory.cli.app import import_app
 from basic_memory.cli.commands.command_utils import run_with_cleanup
 from basic_memory.config import ConfigManager, get_project_config
-from basic_memory.importers.memory_json_importer import MemoryJsonImporter
-from basic_memory.markdown import EntityParser, MarkdownProcessor
-from basic_memory.services.file_service import FileService
 from loguru import logger
 from rich.console import Console
 from rich.panel import Panel
 
+if TYPE_CHECKING:
+    from basic_memory.markdown import MarkdownProcessor
+    from basic_memory.services.file_service import FileService
+
 console = Console()
 
 
 async def get_importer_dependencies() -> Tuple[MarkdownProcessor, FileService]:
     """Get MarkdownProcessor and FileService instances for importers."""
+    # Deferred: the markdown/file-service stack pulls SQLAlchemy and must load
+    # only when an import actually runs, not on every CLI start (#886).
+    from basic_memory.markdown import EntityParser, MarkdownProcessor
+    from basic_memory.services.file_service import FileService
+
     config = get_project_config()
     app_config = ConfigManager().config
     entity_parser = EntityParser(config.home)
@@ -55,6 +64,9 @@ def memory_json(
         markdown_processor, file_service = run_with_cleanup(get_importer_dependencies())
 
         # Create the importer
+        # Deferred: importer stack loads at import-command run time only (#886).
+        from basic_memory.importers.memory_json_importer import MemoryJsonImporter
+
         importer = MemoryJsonImporter(
             config.home, markdown_processor, file_service, project_name=config.name
         )