fix: Smart Transform component fails with ibm watsonx ai language model

[ricofurtado]

Summary by CodeRabbit

• New Features

  • Added new flow querying functions for improved filtering and retrieval capabilities

• Improvements

  • Updated documentation links across starter projects to current documentation pages
  • Enhanced session management with automatic fallback handling in chat components
  • Improved input validation and error handling in chat and output components
  • Updated dependency versions for better compatibility and performance

• Chores

  • Updated package dependencies including security and compatibility patches

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[coderabbitai]

Walkthrough

Updates .secrets.baseline with new secret entries for starter project JSON files, bumps dependency versions in pyproject.toml (cuga and agent-lifecycle-toolkit), adds new async flow retrieval utility functions in src/backend/base/langflow/helpers/flow.py, and updates embedded component code across multiple starter project JSON files with improved session handling, documentation URL changes, and dependency version bumps (particularly FastAPI 0.120.0 → 0.123.0).

Changes

Cohort / File(s)	Summary
Secrets Baseline .secrets.baseline	Added new Hex High Entropy String secret entries for six starter project JSON files (Document Q&A, News Aggregator, Portfolio Website Code Generator, Research Translation Loop, Text Sentiment Analysis, Vector Store RAG) with associated line numbers and hashed values.
Dependency Updates pyproject.toml	Updated cuga from 0.1.10 to ~0.1.11 and agent-lifecycle-toolkit from ~0.4.1 to ~0.4.4.
Flow Helper Utilities src/backend/base/langflow/helpers/flow.py	Added public constant SORT_DISPATCHER mapping sort directions and three new async functions: list_flows_by_flow_folder(), list_flows_by_folder_id(), and get_flow_by_id_or_name() for flexible flow querying with dynamic sorting and filtering; moved Data type import to module scope.
Starter Project Updates: Core Components src/backend/base/langflow/initial_setup/starter_projects/Basic Prompt Chaining.json, Basic Prompting.json, Document Q&A.json, Instagram Copywriter.json, Price Deal Finder.json, SEO Keyword Generator.json, SaaS Pricing.json	Updated ChatInput and ChatOutput components with improved session_id handling (fallback to graph.session_id), changed documentation URLs, added helper methods to ChatOutput (_build_source, _serialize_data, _validate_input, convert_to_string), and updated code_hash metadata.
Starter Project Updates: Advanced Features src/backend/base/langflow/initial_setup/starter_projects/Blog Writer.json, Custom Component Generator.json, Financial Report Parser.json, Image Sentiment Analysis.json, Invoice Summarizer.json, Knowledge Ingestion.json, Knowledge Retrieval.json, Meeting Summary.json, Memory Chatbot.json, Research Agent.json, Research Translation Loop.json	Comprehensive refactoring of component logic including StructuredOutput, Parser, Memory, and Language Model components; added trustcall-based extraction fallback paths, enhanced input validation and error handling, improved source construction, and updated dependency versions and documentation URLs.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~45 minutes

• Flow helper utilities (src/backend/base/langflow/helpers/flow.py): New async query functions with dynamic sorting dispatcher and aliased table joins require careful review of database interaction patterns and error handling.
• Starter project JSON updates: High volume of heterogeneous changes across 15+ files with diverse modifications (component logic refactoring, async/sync method changes, session handling, source construction, provider switching) demand individual file reasoning despite some pattern repetition.
• Dependency version bumps: While straightforward, cascading updates across FastAPI, langchain-core, and langchain-cohere versions should be verified for compatibility.
• Secrets baseline: Addition of multiple baseline entries requires verification that these are legitimate secrets requiring management rather than false positives.

Possibly related PRs

• #10804 — Modifies the same flow.py helper functions and .secrets.baseline entries for starter project JSONs.
• #10867 — Updates ChatOutput/message_response logic in starter project JSONs to preserve session_id handling patterns.
• #10888 — Adds identical SORT_DISPATCHER and flow-listing functions alongside .secrets.baseline and pyproject.toml updates.

Suggested labels

bug

Suggested reviewers

• carlosrcoelho
• ogabrielluiz
• edwinjosechittilappilly

Pre-merge checks and finishing touches

❌ Failed checks (3 warnings)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	You can run @coderabbitai generate docstrings to improve docstring coverage.
Test Quality And Coverage	⚠️ Warning	Test coverage for three new async functions and SORT_DISPATCHER constant is inadequate, primarily containing smoke tests without validating actual behavior.	Add comprehensive tests covering database exception scenarios, null returns, actual sorting verification, UUID conversion errors, SORT_DISPATCHER mappings, invalid order_params, and integration tests with real database fixtures.
Test File Naming And Structure	⚠️ Warning	Pull request makes functional changes to WatsonxAIComponent and LambdaFilterComponent but lacks test coverage for these bug fixes.	Add test_watsonx_stop_sequence.py and test_lambda_filter_logging.py with comprehensive pytest test cases covering positive and negative scenarios.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Test Coverage For New Implementations	✅ Passed	PR includes 14 real unit tests covering three async functions with proper async patterns, mock interactions, and meaningful assertions.
Excessive Mock Usage Warning	✅ Passed	PR #11066 contains no test file modifications; changes are limited to configuration files and helper functions.
Title check	✅ Passed	The PR title clearly identifies a specific bug fix (Smart Transform component failure with IBM Watsonx AI) and accurately reflects the primary change in the changeset.

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch lfoss-2921-smart-transform-component-fails-with-ibm-watsonx-ai-language-model

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[coderabbitai]

Actionable comments posted: 1

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (3)

src/backend/base/langflow/initial_setup/starter_projects/Research Translation Loop.json (1)

405-414: Version update is incomplete and does not achieve stated objective.

FastAPI 0.124.2 (PyPI latest) and 0.124.4 (GitHub latest) are available, but the metadata shows a bump to only 0.123.0. This represents an outdated dependency that falls 2-3 patch versions behind current releases. If the PR's objective is to update dependencies, the fastapi version should be bumped to at least 0.124.0 or newer, not 0.123.0.

src/backend/base/langflow/initial_setup/starter_projects/Meeting Summary.json (1)

3150-3176: Provider options in template don't match embedded component code.

The template options array lists only ["OpenAI", "Anthropic", "Google"] with 3 options_metadata entries, but the embedded component code declares 5 providers (["OpenAI", "Anthropic", "Google", "IBM watsonx.ai", "Ollama"]) with full handler implementations. The template should be updated to include all 5 providers with corresponding metadata entries, or the embedded code should be reduced to match the template. The mismatch creates an inconsistency where users cannot select IBM watsonx.ai or Ollama from the dropdown despite the code supporting them.

src/backend/base/langflow/initial_setup/starter_projects/Instagram Copywriter.json (1)

1044-1120: Remove unused _serialize_data method from ChatOutput.

FastAPI 0.123.0 is a valid release. However, the _serialize_data method defined in the ChatOutput class is not invoked anywhere in message_response() or convert_to_string() and should be removed to eliminate dead code.

♻️ Duplicate comments (9)

src/backend/base/langflow/initial_setup/starter_projects/Price Deal Finder.json (2)

419-495: ChatOutput: same shared implementation as in SaaS Pricing

This ChatOutput block is identical to the implementation already reviewed in SaaS Pricing.json; the previous approval and reasoning apply here as well.

---

1606-1797: AgentComponent: shared provider‑aware agent implementation

This embedded AgentComponent matches the shared version reviewed in *SaaS Pricing.json`; it reuses the same provider‑agnostic LLM wiring and structured‑output logic, so the earlier approval (including the IBM watsonx.ai considerations) applies here too.

src/backend/base/langflow/initial_setup/starter_projects/Knowledge Retrieval.json (1)

235-310: ChatOutput: shared implementation reused here

This ChatOutput definition is the same shared implementation already approved in the other starters (session/source handling, input validation, serialization); no additional concerns specific to this flow.

src/backend/base/langflow/initial_setup/starter_projects/Image Sentiment Analysis.json (3)

237-237: ChatInput starter matches other flows; behavior looks consistent

This ChatInput implementation is the same as in the Invoice Summarizer starter, with proper file normalization and session_id‑based storage gating. The earlier note about optionally guarding self.graph.session_id with an hasattr check applies here as well if you ever expect to run this component without an attached graph.

Also applies to: 285-285

---

511-511: ChatOutput starter aligns with shared implementation; no additional issues spotted

This ChatOutput matches the shared implementation reviewed in the other starter (conversion, validation, session handling, and source metadata). No extra issues here beyond the optional graph‑presence guard already mentioned.

Also applies to: 520-520, 585-585

---

1560-1560: Second LanguageModelComponent block mirrors the first

The second LanguageModelComponent code block is identical to the first one, so all the observations about provider handling and external interactions apply here as well.

src/backend/base/langflow/initial_setup/starter_projects/Document Q&A.json (1)

414-489: ChatOutput here matches the reviewed implementation in Blog Writer

This ChatOutput definition is identical to the one already reviewed in Blog Writer.json (including _build_source, _validate_input, and convert_to_string). The same comments about defensive graph handling apply; otherwise, the implementation looks good.

src/backend/base/langflow/initial_setup/starter_projects/Basic Prompting.json (2)

120-169: ChatInput matches the implementation reviewed in Document Q&A

This ChatInput definition is the same as in Document Q&A.json (including the updated docs URL and file handling). The same suggestion applies to make session_id derivation robust against missing graph:

-        session_id = self.session_id or self.graph.session_id or ""
+        graph = getattr(self, "graph", None)
+        session_id = self.session_id or (graph.session_id if graph else "")

Otherwise, the component looks good.

---

587-662: ChatOutput identical to previously reviewed implementation

This ChatOutput block (including _build_source, _validate_input, convert_to_string, and the new docs URL) is identical to the version already reviewed in Blog Writer.json. The same notes apply:

• Behavior and error handling are good and should work with different model providers, including IBM watsonx.ai.
• Consider guarding self.graph when computing message.session_id and message.flow_id, as suggested there.

No additional issues beyond that.

🧹 Nitpick comments (11)

src/backend/base/langflow/initial_setup/starter_projects/Basic Prompt Chaining.json (1)

709-709: Type annotation mismatch in _build_source method.

The method signature declares source: str | None but the implementation handles objects with model_name or model attributes. Consider updating the type hint:

-def _build_source(self, id_: str | None, display_name: str | None, source: str | None) -> Source:
+def _build_source(self, id_: str | None, display_name: str | None, source: Any) -> Source:

Also note: In convert_to_string, the clean_data parameter is passed to safe_convert only when input_value is a list, but not for single items. If this inconsistency is intentional, consider adding a comment to clarify.

src/backend/base/langflow/helpers/flow.py (1)

102-117: Redundant None check on order_params.

After lines 102-103, order_params is guaranteed to be a dict, so the check on line 114 (if order_params is not None) is always true and can be removed.

     if order_params is None:
         order_params = {"column": "updated_at", "direction": "desc"}

     try:
         async with session_scope() as session:
             uuid_user_id = UUID(user_id) if isinstance(user_id, str) else user_id
             uuid_folder_id = UUID(folder_id) if isinstance(folder_id, str) else folder_id
             stmt = (
                 select(Flow.id, Flow.name, Flow.updated_at)
                 .where(Flow.user_id == uuid_user_id)
                 .where(Flow.folder_id == uuid_folder_id)
             )
-            if order_params is not None:
-                sort_col = getattr(Flow, order_params.get("column", "updated_at"), Flow.updated_at)
-                sort_dir = SORT_DISPATCHER.get(order_params.get("direction", "desc"), desc)
-                stmt = stmt.order_by(sort_dir(sort_col))
+            sort_col = getattr(Flow, order_params.get("column", "updated_at"), Flow.updated_at)
+            sort_dir = SORT_DISPATCHER.get(order_params.get("direction", "desc"), desc)
+            stmt = stmt.order_by(sort_dir(sort_col))

src/backend/base/langflow/initial_setup/starter_projects/Invoice Summarizer.json (2)

308-309: ChatOutput session handling & validation look correct; only minor defensive option

The new ChatOutput implementation (conversion + validation + gated persistence) looks solid and consistent with other starters. The only small thing you might consider is guarding self.graph.session_id the same way you already guard self.graph.flow_id (e.g., via hasattr(self, "graph")) if there is any chance ChatOutput can be used outside a graph context; if not, current code is fine.

Also applies to: 317-317, 383-383

---

864-864: ChatInput files cleanup and session gating are good; same note about graph availability

The updated ChatInput correctly normalizes files to a clean list and only persists messages when a session_id is present. As with ChatOutput, this assumes self.graph is always set when resolving session_id; if ChatInput can ever execute without an attached graph, adding an hasattr(self, "graph") guard around self.graph.session_id would prevent an AttributeError.

Also applies to: 914-914

src/backend/base/langflow/initial_setup/starter_projects/Image Sentiment Analysis.json (1)

1778-1778: StructuredOutputComponent trustcall + LangChain fallback design is sound; double‑check result shapes from get_chat_result

The new StructuredOutputComponent builds a Pydantic list‑wrapper model from output_schema, then:

• Tries a Trustcall‑based extractor first.
• Falls back to llm.with_structured_output(schema) if Trustcall isn’t available or fails.
• Normalizes results into either a list of objects (for Data/DataFrame outputs) or raises when nothing usable is returned.

Overall the flow and error handling look good and should improve robustness with different providers (including IBM watsonx.ai). The only thing to be careful about is that both _extract_output_with_trustcall and _extract_output_with_langchain rely on the concrete shape returned by get_chat_result (list vs. dict with responses, BaseModel vs. plain dict). It would be worth adding/confirming tests that cover:

• Trustcall success and failure paths.
• The LangChain fallback for providers where tool calling/Trustcall isn’t supported.
• Using at least one IBM watsonx.ai model through this component to ensure structured outputs parse as expected.

If those tests pass, the implementation here should be in good shape.

Also applies to: 1849-1849

src/backend/base/langflow/initial_setup/starter_projects/Blog Writer.json (3)

480-555: ChatOutput: behavior is good, but session handling should guard missing graph

The new ChatOutput implementation (conversion, validation, and _build_source) is solid and fixes issues when source is a model object (e.g., non‑OpenAI) by always coercing it to a string before building Source.

One edge case: both message_response implementations use self.graph.session_id without checking whether self.graph is set/non‑None, which can raise AttributeError when the component is used outside a fully wired graph (tests, programmatic usage, tools).

Consider hardening this:

-        message.session_id = self.session_id or self.graph.session_id or ""
-        message.flow_id = self.graph.flow_id if hasattr(self, "graph") else None
+        graph = getattr(self, "graph", None)
+        message.session_id = self.session_id or (graph.session_id if graph else "")
+        message.flow_id = graph.flow_id if graph else None

Same pattern applies to the identical ChatOutput definitions in other starter projects.

---

780-831: ParserComponent: logic is sound; minor config simplification possible

The new ParserComponent correctly handles DataFrame vs Data, structured dicts, and raises clear errors for unsupported types. The Stringify path via convert_to_string and the default‑value handling for missing keys are reasonable.

Two minor points you may want to tighten:

• update_build_config uses self.mode instead of field_value, and if field_value: will always be truthy for "Parser"/"Stringify". You can simplify and make it more explicit:

-        if field_name == "mode":
-            build_config["pattern"]["show"] = self.mode == "Parser"
-            build_config["pattern"]["required"] = self.mode == "Parser"
-            if field_value:
+        if field_name == "mode":
+            is_parser = field_value == "Parser"
+            build_config["pattern"]["show"] = is_parser
+            build_config["pattern"]["required"] = is_parser
+            if is_parser:
                 ...
-            else:
-                build_config.pop("clean_data", None)
+            else:
+                build_config.pop("clean_data", None)

• In convert_to_string, safe_convert(self.input_data or False) will pass False if input_data is falsy, which is probably not intended for Data/DataFrame; you might just drop the or False.

These are non‑blocking and can be deferred.

---

976-1089: URLComponent: implementation is robust; watch for loader exceptions outside RequestException

The new URLComponent adds good validation (ensure_url, validate_url), clear configuration, and structured output with safe conversion and metadata, which improves reliability and downstream use.

One thing to keep in mind: inside fetch_url_contents, only requests.exceptions.RequestException is handled per‑URL; any other exception from RecursiveUrlLoader.load() will bubble to the outer except Exception, aborting the whole operation instead of just skipping the bad URL. If RecursiveUrlLoader can raise non‑RequestException errors (e.g., parsing issues), you may want to broaden the inner except or adjust your logging/continue behavior.

Otherwise this component looks solid.

src/backend/base/langflow/initial_setup/starter_projects/Document Q&A.json (3)

148-199: ChatInput: good file handling; consider safer graph access

The updated ChatInput correctly normalizes files into a list, filters out empty values, and includes them when creating the Message. This should avoid previous None/empty entries issues.

As with ChatOutput, session_id is derived via self.session_id or self.graph.session_id or "". If self.graph is ever unset or None (e.g., in tests or tool calls), this will raise. A more defensive pattern would be:

-        session_id = self.session_id or self.graph.session_id or ""
+        graph = getattr(self, "graph", None)
+        session_id = self.session_id or (graph.session_id if graph else "")

Behavior in normal flows stays the same but becomes safer in edge contexts.

---

1190-1227: FileComponent metadata/dependencies: description clarified; ensure dependency alignment

The File component description is clearer (“Loads and returns the content from uploaded files”), and metadata now declares langchain_core and pydantic alongside lfx. That matches the new use of StructuredTool and BaseModel in the component code.

Please double‑check that:

• langchain_core==0.3.80 and pydantic==2.11.10 are compatible with the versions pinned in pyproject.toml and elsewhere.
• Runtime environments don’t have conflicting pydantic v1/v2 usages.

If everything is already aligned, this metadata change is fine.

---

1286-1615: FileComponent: strong design with Docling isolation; a few edge cases to consider

The new FileComponent is a substantial improvement: tool‑friendly (_get_tools), dynamic description that names files, content‑type validation for images, Docling processing isolated to a subprocess, and clear advanced‑mode behavior. Overall it’s thoughtfully implemented.

A few non‑blocking considerations:

1. Handling advanced_data is None from Docling path

   In process_files, advanced path:

   advanced_data: Data | None = self._process_docling_in_subprocess(file_path)
   payload = getattr(advanced_data, "data", {}) or {}
   doc_rows = payload.get("doc")
   ...
   else:
       final_return.extend(self.rollup_data(file_list, [advanced_data]))

   If _process_docling_in_subprocess returns None (e.g., empty file_path or unexpected failure), advanced_data is None and gets passed into rollup_data, which may not expect None. Consider short‑circuiting when advanced_data is falsy and logging/raising instead of rolling it up.

2. Standard path concurrency flag semantics

   You’ve marked use_multithreading as deprecated and documented that “Processing Concurrency > 1” enables multithreading, but the code still gates on use_multithreading:

   concurrency = 1 if not self.use_multithreading else max(1, self.concurrency_multithreading)

   If the intent is “any concurrency_multithreading > 1 enables parallelism”, you could simplify to:

   concurrency = max(1, self.concurrency_multithreading)

   and ignore use_multithreading, or auto‑derive it from the concurrency value.

3. Docling compatibility and formats

   _is_docling_compatible includes a wide set of extensions (including some structured ones). Combined with advanced_mode gates, this is probably intentional, but it means that if someone somehow forces advanced_mode=True for a .csv/.json file (e.g., via API), Docling will be tried and any failure will only surface via the Data(error=...) path. That’s acceptable, just be aware of the behavior.

4. Subprocess path safety

   You already sanitize args["file_path"] for shell metacharacters and pass config via stdin to python -c, which is good. Since you’re not invoking a shell and not interpolating the path into the command, the risk is low; the extra filter is mainly defense‑in‑depth and looks fine.

None of these are blockers; the component should work well as written.

📜 Review details

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 841a133 and 942174a.

⛔ Files ignored due to path filters (2)

• src/frontend/package-lock.json is excluded by !**/package-lock.json
• uv.lock is excluded by !**/*.lock

📒 Files selected for processing (21)

• .secrets.baseline (1 hunks)
• pyproject.toml (1 hunks)
• src/backend/base/langflow/helpers/flow.py (3 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Basic Prompt Chaining.json (5 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Basic Prompting.json (5 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Blog Writer.json (9 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Custom Component Generator.json (7 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Document Q&A.json (8 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Financial Report Parser.json (9 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Image Sentiment Analysis.json (7 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Instagram Copywriter.json (9 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Invoice Summarizer.json (7 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Knowledge Ingestion.json (5 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Knowledge Retrieval.json (6 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Meeting Summary.json (13 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Memory Chatbot.json (7 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Price Deal Finder.json (7 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Research Agent.json (7 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/Research Translation Loop.json (11 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/SEO Keyword Generator.json (3 hunks)
• src/backend/base/langflow/initial_setup/starter_projects/SaaS Pricing.json (7 hunks)

🧰 Additional context used

📓 Path-based instructions (2)

{pyproject.toml,uv.lock}

📄 CodeRabbit inference engine (.cursor/rules/backend_development.mdc)

Use uv (>=0.4) for Python dependency management in backend development

Files:

• pyproject.toml

src/backend/**/*.py

📄 CodeRabbit inference engine (.cursor/rules/backend_development.mdc)

src/backend/**/*.py: Use FastAPI async patterns with await for async operations in component execution methods
Use asyncio.create_task() for background tasks and implement proper cleanup with try/except for asyncio.CancelledError
Use queue.put_nowait() for non-blocking queue operations and asyncio.wait_for() with timeouts for controlled get operations

Files:

• src/backend/base/langflow/helpers/flow.py

🧠 Learnings (4)

📓 Common learnings

Learnt from: CR
Repo: langflow-ai/langflow PR: 0
File: .cursor/rules/backend_development.mdc:0-0
Timestamp: 2025-11-24T19:46:09.104Z
Learning: Applies to src/backend/base/langflow/components/**/*.py : Add new components to the appropriate subdirectory under `src/backend/base/langflow/components/` (agents/, data/, embeddings/, input_output/, models/, processing/, prompts/, tools/, or vectorstores/)

📚 Learning: 2025-11-24T19:46:09.104Z

Learnt from: CR
Repo: langflow-ai/langflow PR: 0
File: .cursor/rules/backend_development.mdc:0-0
Timestamp: 2025-11-24T19:46:09.104Z
Learning: Applies to src/backend/base/langflow/components/**/__init__.py : Update `__init__.py` with alphabetically sorted imports when adding new components

Applied to files:

• src/backend/base/langflow/initial_setup/starter_projects/Custom Component Generator.json
• src/backend/base/langflow/helpers/flow.py

📚 Learning: 2025-06-26T19:43:18.260Z

Learnt from: ogabrielluiz
Repo: langflow-ai/langflow PR: 0
File: :0-0
Timestamp: 2025-06-26T19:43:18.260Z
Learning: In langflow custom components, the `module_name` parameter is now propagated through template building functions to add module metadata and code hashes to frontend nodes for better component tracking and debugging.

Applied to files:

• src/backend/base/langflow/initial_setup/starter_projects/Research Translation Loop.json
• src/backend/base/langflow/initial_setup/starter_projects/Knowledge Ingestion.json

📚 Learning: 2025-11-24T19:47:28.997Z

Learnt from: CR
Repo: langflow-ai/langflow PR: 0
File: .cursor/rules/testing.mdc:0-0
Timestamp: 2025-11-24T19:47:28.997Z
Learning: Applies to src/backend/tests/**/*.py : Use predefined JSON flows and utility functions from `tests.unit.build_utils` (create_flow, build_flow, get_build_events, consume_and_assert_stream) for flow execution testing

Applied to files:

• src/backend/base/langflow/helpers/flow.py

🧬 Code graph analysis (1)

src/backend/base/langflow/helpers/flow.py (3)

src/lfx/src/lfx/schema/data.py (1)

• Data (26-288)

src/lfx/src/lfx/helpers/flow.py (3)

• list_flows_by_flow_folder (88-131)
• list_flows_by_folder_id (134-169)
• get_flow_by_id_or_name (172-202)

src/backend/tests/conftest.py (1)

• flow (549-569)

🔇 Additional comments (48)

src/backend/base/langflow/initial_setup/starter_projects/Financial Report Parser.json (3)

1163-1163: Good fallback mechanism for model compatibility.

The implementation correctly handles the IBM watsonx.ai compatibility issue by first attempting extraction with trustcall, then falling back to LangChain's with_structured_output when trustcall fails. The warning message appropriately explains that some models don't support tool calling, which aligns with the PR objective.

---

229-229: Consistent session ID fallback pattern.

The session handling logic self.session_id or self.graph.session_id or "" provides a robust fallback chain. This pattern is consistently applied in both ChatOutput and ChatInput components.

---

160-163: FastAPI version 0.123.0 is available and valid.

The dependency version bump is not an issue—this version exists and is established in the release timeline.

src/backend/base/langflow/initial_setup/starter_projects/Basic Prompt Chaining.json (4)

364-376: LGTM!

Metadata updates (code_hash, dependencies, module path) correctly reflect the updated ChatInput component code.

---

413-413: Improved session ID handling looks good.

The fallback chain self.session_id or self.graph.session_id or "" ensures consistent session resolution across the flow. The async Message.create() pattern and file filtering logic are correct.

---

1346-1361: Provider options in template don't include IBM watsonx.ai or Ollama.

The template's options array only lists ["OpenAI", "Anthropic", "Google"], but the embedded code supports 5 providers including "IBM watsonx.ai" and "Ollama". Given the PR title mentions fixing IBM watsonx.ai issues, verify this mismatch is intentional and that users can still select IBM watsonx.ai from the dropdown in the UI.

The same pattern appears in the other two LanguageModelComponent nodes (lines 1668-1693, 1989-2014).

---

643-645: FastAPI 0.123.0 is a valid PyPI release. Consider whether pinning to a more recent 0.123.x patch version (0.123.9 or 0.123.10) may be preferred for bug fixes and stability.

src/backend/base/langflow/initial_setup/starter_projects/Custom Component Generator.json (3)

2240-2250: FastAPI 0.123.0 dependency bump is safe to proceed with.

The upgrade from FastAPI 0.120.0 to 0.123.0 does not introduce breaking changes affecting fastapi.encoders.jsonable_encoder or standard FastAPI imports. Changes between these versions focus on dependency handling improvements and Pydantic v1 deprecation notices, not on core API modifications.

---

302-302: Embedded Python code syntax validates correctly.

The three embedded Python code blocks have been verified and all pass Python syntax validation:

• Memory Component: 10,363 characters, 12 import statements, 4 async methods
• ChatInput Component: 3,483 characters, 6 import statements, 1 async method
• ChatOutput Component: 7,003 characters, 13 import statements, 1 async method

All imports are valid, method signatures are compatible with dependencies, and Session ID handling logic is properly implemented.

---

240-240: Code hash metadata is correctly updated and consistent.

All three code_hash values match their embedded code blocks:

• Memory Component (line 240): efd064ef48ff ✓
• ChatInput (line 1960): 7a26c54d89ed ✓
• ChatOutput (line 2240): cae45e2d53f6 ✓

src/backend/base/langflow/initial_setup/starter_projects/Knowledge Ingestion.json (3)

92-92: SplitTextComponent updates look consistent.

The code_hash update and documentation URL change to https://docs.langflow.org/split-text are consistent with the embedded code modifications.

Also applies to: 181-181

---

356-356: URLComponent updates look consistent.

The code_hash update and documentation URL change to https://docs.langflow.org/url are consistent with the embedded code modifications.

Also applies to: 467-467

---

773-773: Dependency version bump for langchain_cohere.

The update from version 0.3.3 to 0.3.5 aligns with the broader dependency updates in this PR.

.secrets.baseline (1)

768-883: Secrets baseline updates for starter project code hashes.

These additions document known false positives for Hex High Entropy String detections in starter project JSON files. The flagged values are code_hash identifiers (12-character hex strings) generated for component tracking, not actual secrets. All entries correctly have "is_secret": false.

src/backend/base/langflow/initial_setup/starter_projects/SEO Keyword Generator.json (2)

639-639: ChatOutput session handling improvements.

The updated message_response method includes improved session_id resolution with a fallback chain:

message.session_id = self.session_id or self.graph.session_id or ""

The guard for Message reuse (not self.is_connected_to_chat_input()) ensures proper message creation when connected to a ChatInput component. The documentation URL update to chat-input-and-output consolidates the reference.

---

565-574: Metadata updates for ChatOutput component.

The code_hash and fastapi dependency version bump (0.120.0 → 0.123.0) are consistent with the embedded code changes.

pyproject.toml (1)

138-139: Dependency version updates are reasonable.

The version bumps for cuga (from ~=0.1.11) and agent-lifecycle-toolkit (to ~=0.4.4) are minor updates. The change in cuga from exact pinning (==) to compatible release (~=) allows patch-level updates within the 0.1.x series, which is appropriate for minor versions.

src/backend/base/langflow/initial_setup/starter_projects/Research Translation Loop.json (5)

480-480: ChatOutput embedded code looks correct.

The embedded ChatOutput component code includes proper session handling with fallback logic (self.session_id or self.graph.session_id or ""), source building with model attribute detection, and input validation. The implementation follows established patterns.

---

687-736: ChatInput embedded code properly handles session context.

The ChatInput component correctly:

• Handles file input normalization
• Falls back to graph.session_id when session_id is not provided
• Uses Message.create() factory method for async message creation
• Properly filters None/empty file values

---

987-1036: ParserComponent code update looks correct.

The ParserComponent properly handles both "Parser" and "Stringify" modes, with appropriate type validation for DataFrame and Data inputs. The update_build_config method correctly toggles the pattern field visibility based on mode selection.

---

1171-1237: LoopComponent code correctly handles Message to Data conversion.

The LoopComponent:

• Properly validates input data types (DataFrame, Data, Message, list)
• Converts Message objects to Data for consistent processing
• Manages loop state via context (ctx) with proper initialization checks
• Updates run dependencies for loop iteration

---

1624-1699: TypeConverterComponent code is well-structured.

The TypeConverterComponent provides clean conversion functions between Message, Data, and DataFrame types with proper auto-parsing support for JSON/CSV content. The update_outputs method correctly manages dynamic output display based on selected output type.

src/backend/base/langflow/initial_setup/starter_projects/Memory Chatbot.json (3)

151-199: ChatInput component metadata and code updates are consistent.

The ChatInput component correctly handles:

• Session ID fallback chain (self.session_id or self.graph.session_id or "")
• Async message creation with Message.create()
• File normalization and filtering
• Context ID for memory layering

---

426-500: ChatOutput component properly handles source attribution.

The _build_source method correctly handles different source object types (checking for model_name and model attributes), and the message_response method properly sets message properties including session/context IDs and flow ID.

---

902-966: Memory component session_id handling is correct and already consistent with other chat components.

The concern about missing fallback logic is unfounded. By default, if the Session ID value is empty, it is set to the same value as the Flow ID, and all components in the flow automatically use this session_id value. The Memory component's info text confirms this: "If empty, the current session ID parameter will be used." This behavior is managed by the framework, not by individual component code. ChatInput and ChatOutput components work the same way—they do not implement explicit fallback logic because the framework handles session_id resolution uniformly across all components. The empty default for session_id in the Memory component is intentional and correct.

src/backend/base/langflow/initial_setup/starter_projects/Research Agent.json (3)

480-528: Updated ChatInput component with improved session handling.

The session ID resolution now correctly falls back through multiple sources: self.session_id or self.graph.session_id or "". This ensures robust session handling when the component-level session ID is not explicitly set.

---

2579-2770: Agent component updated with langchain_core 0.3.80 and improved exception handling.

The Agent component code shows proper exception handling patterns and structured output validation. The dependency version bump for langchain_core aligns with the coordinated update.

---

1644-1720: ChatOutput component updated with FastAPI 0.123.0.

The _build_source helper method properly handles different model object types (checking for model_name, then model, then falling back to str()). The code is compatible with the specified FastAPI version.

Note: FastAPI 0.124.4 is now available as the latest version. Consider upgrading if this is part of a broader dependency update initiative.

src/backend/base/langflow/helpers/flow.py (4)

9-10: New SQLAlchemy imports for aliased joins and sorting.

The imports for aliased, asc, and desc support the new query patterns for flow listing. This is a clean approach for dynamic ordering.

---

30-33: SORT_DISPATCHER provides clean dynamic ordering.

Using a dictionary dispatcher for sort direction is a good pattern that avoids repetitive if-else chains and provides safe defaults.

---

52-89: New list_flows_by_flow_folder retrieves flows in the same folder as a given flow.

The self-join via aliased(Flow) correctly identifies the folder of the specified flow and returns sibling flows. The noqa: B006 for the mutable default dict is acceptable since it's only read, not mutated. The _mapping access (line 86) is a SQLAlchemy Row attribute - consider using ._asdict() for cleaner conversion, though this is minor.

---

126-163: New get_flow_by_id_or_name correctly prioritizes flow_id over flow_name.

The logic properly handles both lookup methods and converts string IDs to UUIDs when needed. The implementation aligns with the lfx stub interface shown in the relevant code snippets.

One minor observation: the check on lines 148-150 is defensive but redundant since lines 142-147 guarantee attr and val are set if we pass the validation on line 135-137. However, this defensive coding is acceptable.

src/backend/base/langflow/initial_setup/starter_projects/SaaS Pricing.json (3)

371-448: ChatOutput: session/source handling and input validation look solid

The updated ChatOutput implementation (validated inputs, convert_to_string, session ID fallback, and _build_source usage) is consistent and robust for Data/DataFrame/Message inputs and aligns with the shared component version used elsewhere. No issues from this starter embedding.

---

689-742: CalculatorComponent: safe AST evaluation with clear error handling

The calculator implementation safely restricts allowed AST node types and operators, handles legacy ast.Num, and returns a cleanly formatted numeric result or structured error in Data. This looks correct and appropriate for a starter tool.

---

868-1059: AgentComponent: provider‑agnostic wiring suitable for IBM watsonx.ai

The embedded AgentComponent correctly delegates provider‑specific details to MODEL_PROVIDERS_DICT/MODEL_PROVIDERS_LIST, filters out json_mode from OpenAI, and centralizes LLM construction in _build_llm_model/get_llm. This should work uniformly for providers like IBM watsonx.ai so long as their entries are present in MODEL_PROVIDERS_DICT, which matches the intent of the PR.

Please double‑check that the IBM watsonx.ai provider entry in MODEL_PROVIDERS_DICT exposes the expected component_class, inputs, and prefix so this generic logic wires it correctly.

src/backend/base/langflow/initial_setup/starter_projects/Price Deal Finder.json (1)

140-191: ChatInput: improved file handling and session resolution

The new ChatInput.message_response correctly normalizes files, ignores empty/None entries, derives session_id with a sensible fallback to the graph session, and only persists when a non‑empty session is present. This is a clear improvement with no obvious regressions.

src/backend/base/langflow/initial_setup/starter_projects/Knowledge Retrieval.json (2)

109-159: TextInputComponent: docs URL and implementation look correct

The TextInputComponent wrapper cleanly maps input_value into a Message and updates the documentation URL to the new text-input-and-output page. No issues here.

---

514-609: KnowledgeRetrievalComponent: end‑to‑end retrieval pipeline is consistent and safe

The new KnowledgeRetrievalComponent correctly:

• Resolves the per‑user knowledge base path and validates user/session context.
• Loads and decrypts embedding metadata, building an appropriate embedding model for OpenAI/HuggingFace/Cohere.
• Uses Chroma similarity search (with and without scores) and optionally enriches results with stored embeddings keyed by _id.
• Wraps rows into Data instances and returns a DataFrame(data=data_list) compatible with the existing Data/DataFrame schemas.

This wiring looks coherent and should behave well for Cohere given the explicit API‑key enforcement and the bumped langchain_cohere version.

It would still be good to run a quick manual retrieval against a Cohere‑backed knowledge base on this branch to confirm there are no runtime issues (e.g., API‑parameter mismatches) with langchain_cohere==0.3.5.

src/backend/base/langflow/initial_setup/starter_projects/Invoice Summarizer.json (1)

1154-1160: AgentComponent provider plumbing (incl. IBM/Ollama) looks coherent; rely on tests for build_config edge cases

The reworked AgentComponent uses MODEL_PROVIDERS_DICT / MODEL_PROVIDERS_LIST consistently to inject and prune provider‑specific fields, and the update_build_config logic for agent_llm/dynamic fields looks internally consistent (including the connect_other_models path and required‑keys check). Given the branching and external dependency on per‑provider component classes, it would be good to exercise this with tests that:

• Switch agent_llm across all providers (including IBM watsonx.ai / Ollama) and back.
• Verify build_config still contains the expected default keys and that provider‑specific fields are added/removed as intended.

This looks correct, but those tests will guard against regressions as the provider matrix evolves.

Also applies to: 1345-1345

src/backend/base/langflow/initial_setup/starter_projects/Image Sentiment Analysis.json (1)

1235-1235: LanguageModelComponent IBM/Ollama integration looks reasonable; please validate provider flows against real backends

The updated LanguageModelComponent cleanly wires in IBM watsonx.ai and Ollama: build_model enforces provider‑specific required fields and instantiates ChatWatsonx / ChatOllama, while update_build_config dynamically updates model options using fetch_ibm_models, get_ollama_models, and URL validation. The control flow and fallbacks (defaults + warnings on failures) look coherent, so I don’t see an obvious correctness bug in the component itself. Given the reliance on external HTTP calls and third‑party client libraries, I’d strongly recommend:

• Exercising provider switches (OpenAI/Anthropic/Google/IBM/Ollama) via tests or manual runs.
• Verifying that IBM model list fetching and Ollama model discovery behave as expected in your target environments (and degrade gracefully when endpoints are unreachable).

src/backend/base/langflow/initial_setup/starter_projects/Blog Writer.json (1)

353-427: TextInputComponent update looks correct and self‑contained

The updated TextInputComponent logic and docs URL are straightforward; it simply wraps input_value into a Message without side effects. No issues from a correctness or UX standpoint.

src/backend/base/langflow/initial_setup/starter_projects/Meeting Summary.json (4)

718-718: Embedded ChatOutput code updates are well-structured.

The updated component code includes:

• New _build_source helper for constructing Source objects with proper null handling
• Enhanced session handling with fallback chain: self.session_id or self.graph.session_id or ""
• Improved input validation via _validate_input method

The same code is consistently applied to all three ChatOutput nodes in this flow.

---

1656-1719: Memory component updates align with the session handling improvements.

The Memory component code includes:

• Proper async methods for store_message and retrieve_messages
• Consistent lfx imports matching other updated components
• Mode-based dynamic output configuration

---

2018-2067: ChatInput component updates are consistent with ChatOutput session handling.

The ChatInput component now uses the same session ID fallback pattern:

session_id = self.session_id or self.graph.session_id or ""

This ensures consistent session handling across input and output components in the flow.

---

642-660: Metadata and dependency version updates are consistent.

The code_hash update and FastAPI version bump from 0.120.0 to 0.123.0 are correctly applied, with 0.123.0 being a valid released version in the 0.123.x series. The changes align with the updated embedded component code that includes enhanced session handling and the new _build_source helper method.

src/backend/base/langflow/initial_setup/starter_projects/Instagram Copywriter.json (3)

318-369: LGTM - Improved session handling in ChatInput.

The session_id fallback chain (self.session_id or self.graph.session_id or "") provides robust handling when no explicit session ID is configured. The async Message.create() pattern is appropriate.

---

760-810: LGTM - TextInput component with updated documentation URL.

The component is straightforward and the documentation URL update aligns with the project's documentation structure.

---

1965-2174: Agent component supports multiple LLM providers including IBM watsonx.ai.

The embedded Agent code includes IBM watsonx.ai as one of the supported MODEL_PROVIDERS alongside OpenAI, Anthropic, Google, and Ollama. The implementation shows proper multi-provider handling with dynamic build_config updates and schema validation for structured outputs. Async error handling is appropriate with specific exception types.

The code is well-structured with proper async/await patterns and comprehensive error handling. The component properly integrates memory management via MemoryComponent and tool calling functionality.

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Bug: Missing clean_data parameter and incorrect fallback in convert_to_string.

In the convert_to_string method, the else branch has an issue:

result = safe_convert(self.input_data or False)

Two problems:

1. self.input_data or False converts falsy inputs (like empty string/list) to boolean False, which is likely unintended
2. The clean_data parameter is missing, unlike the list branch which correctly passes clean_data=self.clean_data or False

Apply this fix within the embedded code string:

-        result = safe_convert(self.input_data or False)
+        result = safe_convert(self.input_data, clean_data=self.clean_data or False)

🤖 Prompt for AI Agents

In src/backend/base/langflow/initial_setup/starter_projects/Financial Report
Parser.json around line 1437, the convert_to_string else branch incorrectly
calls safe_convert(self.input_data or False) and omits the clean_data argument;
change it to call safe_convert(self.input_data, clean_data=self.clean_data or
False) (i.e., pass the actual input_data without coercing falsy values to False
and include the same clean_data parameter used in the list branch) so behavior
and parameters mirror the list handling.