fix: make lancedb optional dependency under [memory] extras

[NIK-TIGER-BILL]

Description

Moves lancedb>=0.29.2 from core dependencies to [project.optional-dependencies] memory group. This unblocks installation on Intel Macs where lancedb has no x86_64 wheels (see #5327).

Changes

• pyproject.toml: Removed lancedb>=0.29.2 from dependencies, added [memory] optional dependency group.
• lancedb_storage.py: Wrapped import lancedb in a try/except block. LanceDBStorage.__init__ now raises a clear ImportError pointing users to pip install crewai[memory].

Impact

• pip install crewai no longer requires lancedb — works on Intel Macs.
• Users who need memory features install with pip install crewai[memory].
• Existing Apple Silicon and Linux users are unaffected.

Related Issue

Fixes #5327

Summary by CodeRabbit

Release Notes

• Documentation

  • Clarified default flow persistence file name and platform-specific storage locations
  • Updated persistence example to demonstrate proper state resumption with explicit ID passing

• Bug Fixes

  • Fixed tool execution to correctly mark results as errors when tool exceptions occur

• Chores

  • Improved optional dependency handling with clearer installation prompts for memory storage features

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR addresses a critical installation blocker by making lancedb an optional dependency, refines tool execution semantics to prevent incorrect result flags on errors, and clarifies flow persistence documentation with platform-specific storage details and corrected state resumption examples.

Changes

crewAI Library Improvements

Layer / File(s)	Summary
Optional lancedb dependency and lazy loading lib/crewai/pyproject.toml, lib/crewai/src/crewai/memory/storage/lancedb_storage.py	Moves lancedb from mandatory core dependency to optional [memory] extras group. LanceDBStorage now uses lazy import with try/except, raising an informative ImportError with install hint (pip install crewai[memory]) when the dependency is unavailable.
Tool error handling refinement lib/crewai/src/crewai/utilities/agent_utils.py, lib/crewai/tests/utilities/test_agent_utils.py	execute_single_native_tool_call now sets result_as_answer only when the tool indicates result_as_answer AND no error event was emitted. New test verifies that tool errors with result_as_answer=True return result_as_answer=False with error details.
Flow persistence documentation docs/en/concepts/flows.mdx, docs/en/guides/flows/mastering-flow-state.mdx	Documented SQLite backend filename (flow_states.db), platform-specific storage paths (Linux, macOS, Windows), default project-name behavior, and CREWAI_STORAGE_DIR override. Updated state resumption example to explicitly pass flow1.state.id into kickoff.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• crewAIInc/crewAI#5779: Related to the updated flow state example; that PR discusses deprecation of inputs.id hydration in favor of restoreFromStateId.

Suggested labels

size/M

Suggested reviewers

• greysonlalonde
• vinibrsl

Poem

🐰 A lancedb made optional, Intel Macs now sing,
Tool errors caught with care, no false flags take wing,
Persistence paths revealed for all to see,
State flows resume with explicit ID—crispy! 🌰✨

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (1 warning, 1 inconclusive)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 40.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.
Out of Scope Changes check	❓ Inconclusive	Documentation and test changes appear tangential to the core lancedb dependency fix; clarification needed on whether updating flow persistence docs and tool error handling tests are necessary scope for this PR.	Confirm whether documentation updates (flows.mdx, mastering-flow-state.mdx) and the tool error handling test are intentional dependencies of the lancedb optional-dependency fix or should be in separate PRs.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'fix: make lancedb optional dependency under [memory] extras' accurately summarizes the main objective of moving lancedb from core dependencies to optional dependencies.
Linked Issues check	✅ Passed	The pull request fully addresses issue #5327 by making lancedb an optional dependency under [memory] extras, ensuring crewai is installable on Intel Macs while allowing opt-in for memory features.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

⚔️ Resolve merge conflicts

• Resolve merge conflict in branch fix-lancedb-optional-5327

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@lib/crewai/src/crewai/utilities/agent_utils.py`:
- Around line 1595-1600: The current is_result_as_answer flag only checks
original_tool.result_as_answer and error_event_emitted, allowing tools marked
result_as_answer to be treated as final even when the tool was blocked or not
found; modify the condition in the is_result_as_answer assignment to also
require overall successful execution by negating the relevant failure flags —
e.g., include checks like not tool_blocked and not tool_not_found (or a single
execution_succeeded flag if present) so the final expression becomes something
like: original_tool and hasattr(original_tool, "result_as_answer") and
original_tool.result_as_answer and not error_event_emitted and not tool_blocked
and not tool_not_found.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c233a85e-0765-4c49-ae66-b064e9976853

📥 Commits

Reviewing files that changed from the base of the PR and between c36827b and cec692d.

📒 Files selected for processing (6)

• docs/en/concepts/flows.mdx
• docs/en/guides/flows/mastering-flow-state.mdx
• lib/crewai/pyproject.toml
• lib/crewai/src/crewai/memory/storage/lancedb_storage.py
• lib/crewai/src/crewai/utilities/agent_utils.py
• lib/crewai/tests/utilities/test_agent_utils.py

[coderabbitai]

⚠️ Potential issue | 🟠 Major | ⚡ Quick win

Guard result_as_answer on overall execution success, not only emitted error events.

At Line 1599, tool-blocked and tool-not-found paths still allow result_as_answer=True for tools configured with result_as_answer, which can incorrectly finalize with an error/block message.

Proposed fix

+    execution_failed = hook_blocked or error_event_emitted or (
+        not from_cache and func_name not in available_functions
+    )
+
     is_result_as_answer = bool(
         original_tool
         and hasattr(original_tool, "result_as_answer")
         and original_tool.result_as_answer
-        and not error_event_emitted
+        and not execution_failed
     )

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@lib/crewai/src/crewai/utilities/agent_utils.py` around lines 1595 - 1600, The
current is_result_as_answer flag only checks original_tool.result_as_answer and
error_event_emitted, allowing tools marked result_as_answer to be treated as
final even when the tool was blocked or not found; modify the condition in the
is_result_as_answer assignment to also require overall successful execution by
negating the relevant failure flags — e.g., include checks like not tool_blocked
and not tool_not_found (or a single execution_succeeded flag if present) so the
final expression becomes something like: original_tool and
hasattr(original_tool, "result_as_answer") and original_tool.result_as_answer
and not error_event_emitted and not tool_blocked and not tool_not_found.