feat(agentic): Langflow Assistant improvements — flow building, model portability, and robustness

[Cristhianzl]

Summary

A broad reliability + capability pass on the Langflow Assistant: it now builds
loop and conditional flows correctly, works across every configured model provider
(fixing silent failures on WatsonX, Ollama, and reasoning models like gpt-5.6),
surfaces recovered errors instead of hiding them, and gives the user session-level
control over memory and step budget. Targets release-1.11.0.

Summary by CodeRabbit

• New Features

  • Added starter-template discovery and application in Assistant workflows.
  • Added loop and branching support with improved flow validation.
  • Added headless Assistant execution and consolidated HTTP MCP access.
  • Added live progress updates for canvas-building actions.
  • Added restore points with the ability to revert Assistant-applied edits.
  • Added /history controls for limiting conversation context.
  • Added structured error details and model fallback/remediation notices.
  • Increased Assistant iteration capacity for complex builds.

• Bug Fixes

  • Improved model portability, streaming reliability, tracing resilience, and canvas edge preservation.
  • Prevented malformed flow updates from modifying the canvas.
  • Expanded localized messaging across supported languages.

• Documentation

  • Updated Assistant and MCP feature documentation with current API behavior and workflow guidance.

[coderabbitai]

Important

Review skipped

Auto incremental reviews are disabled on this repository.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 01527e6a-6af7-468f-9fef-e988aeeb373b

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

Walkthrough

The change expands the agentic assistant with headless execution, HTTP MCP access, loop-aware flow building and validation, bounded history, progress events, restore/revert behavior, structured errors, template tools, atomic canvas updates, and frontend localization with extensive regression coverage.

Changes

Agentic assistant and MCP

Layer / File(s)	Summary
Assistant execution and API contracts src/backend/base/langflow/agentic/api/*, src/backend/base/langflow/agentic/services/*	Adds the headless SSE route, history limits, restore-point metadata, structured errors, bounded history injection, and live provider entries.
MCP transport and tools src/backend/base/langflow/api/v1/agentic_mcp.py, src/backend/base/langflow/agentic/mcp/*, src/lfx/src/lfx/mcp/*	Adds authenticated HTTP MCP delegation, assistant execution through the headless route, progress forwarding, and starter-template tools.
Loop-aware flow construction src/lfx/src/lfx/graph/flow_builder/*, src/lfx/src/lfx/mcp/registry.py	Adds loop feedback handles, loop input discovery, dropdown-output synchronization, conversion hints, and tool-start instrumentation.
Frontend assistant state and controls src/frontend/src/components/core/assistantPanel/*, src/frontend/src/controllers/API/queries/agentic/*, src/frontend/src/stores/flowStore.ts	Adds tool-progress rows, structured error details, restore/revert actions, history commands, atomic node-edge updates, and related SSE types.
Validation and regression coverage src/backend/tests/*, src/lfx/tests/*, src/frontend/src/**/__tests__/*	Adds coverage for loop validation, MCP routes, templates, progress events, restore points, error details, history bounds, canvas updates, and frontend lifecycle behavior.

Estimated code review effort: 5 (Critical) | ~120 minutes

Possibly related PRs

• langflow-ai/langflow#12981: Both changes modify the flow executor’s streaming event path.
• langflow-ai/langflow#13107: Both changes modify Langfuse client lifecycle and test-reset behavior.
• langflow-ai/langflow#13530: Both changes modify agentic API routing and assistant endpoint exposure.

Suggested reviewers: ogabrielluiz, adam-aghili, jkavia

🚥 Pre-merge checks | ✅ 8 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 19.52% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (8 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Test Coverage For New Implementations	✅ Passed	The PR adds convention-named tests across backend, frontend, and LFX, and sampled files contain real assertions for loops, restore points, tool_start, and error details.
Test Quality And Coverage	✅ Passed	Broad behavior-focused coverage exists for loops, history, restore, tool_start, API success/error paths, and frontend revert/error states, using proper pytest/RTL async patterns.
Test File Naming And Structure	✅ Passed	PASS: Changed test files match repo conventions (test_*.py backend, *.test.ts/tsx frontend) and representative suites use descriptive pytest/Jest cases with setup/teardown.
Excessive Mock Usage Warning	✅ Passed	Mocks are mostly confined to boundary tests (streaming, preload, hook plumbing); core loop/validation tests use real fixtures and show 0 mock hits, so usage isn’t excessive.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately summarizes the main assistant improvements around flow building, model portability, and robustness.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch cz/assistant-improvements

---

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.}

[github-actions]

✅ Test Coverage Advisor

No source changes detected without accompanying tests. Thanks for keeping coverage up! 🎉

Advisory check only — never blocks merge.

[codecov]

Codecov Report

❌ Patch coverage is 82.47768% with 314 lines in your changes missing coverage. Please review.
✅ Project coverage is 61.53%. Comparing base (597d05f) to head (db6e893).
⚠️ Report is 36 commits behind head on release-1.12.0.

Files with missing lines	Patch %	Lines
...e/assistantPanel/hooks/use-assistant-chat.types.ts	0.00%	58 Missing ⚠️
...ntend/src/controllers/API/queries/agentic/types.ts	0.00%	40 Missing ⚠️
...nents/core/assistantPanel/assistant-panel.types.ts	0.00%	39 Missing ⚠️
src/lfx/src/lfx/mcp/server.py	10.71%	25 Missing ⚠️
...components/core/assistantPanel/assistant-panel.tsx	35.13%	24 Missing ⚠️
...nents/core/assistantPanel/hooks/history-storage.ts	81.31%	17 Missing ⚠️
src/frontend/src/types/zustand/flow/index.ts	0.00%	9 Missing ⚠️
src/lfx/src/lfx/graph/flow_builder/connect.py	88.15%	5 Missing and 4 partials ⚠️
...rc/lfx/src/lfx/mcp/flow_builder_tools/run_tools.py	30.76%	7 Missing and 2 partials ⚠️
...ssistantPanel/components/assistant-build-tasks.tsx	89.33%	8 Missing ⚠️
... and 23 more

Additional details and impacted files

@@                Coverage Diff                 @@
##           release-1.12.0   #14007      +/-   ##
==================================================
+ Coverage           61.47%   61.53%   +0.05%     
==================================================
  Files                2339     2353      +14     
  Lines              237145   238537    +1392     
  Branches            35410    33490    -1920     
==================================================
+ Hits               145787   146779     +992     
- Misses              89561    89995     +434     
+ Partials             1797     1763      -34     

Flag	Coverage Δ
backend	69.34% <92.77%> (+0.39%)	⬆️
frontend	59.75% <79.48%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/backend/base/langflow/agentic/api/schemas.py	100.00% <100.00%> (ø)
...e/langflow/agentic/flows/flow_builder_assistant.py	100.00% <ø> (ø)
...end/base/langflow/agentic/helpers/code_security.py	91.11% <100.00%> (-0.82%)	⬇️
...ase/langflow/agentic/helpers/input_sanitization.py	100.00% <100.00%> (ø)
src/backend/base/langflow/agentic/mcp/server.py	64.44% <100.00%> (+8.88%)	⬆️
...e/langflow/agentic/services/conversation_buffer.py	96.72% <100.00%> (-3.28%)	⬇️
...nd/base/langflow/agentic/services/flow_executor.py	94.53% <100.00%> (-0.60%)	⬇️
...angflow/agentic/services/helpers/event_consumer.py	94.44% <100.00%> (-5.56%)	⬇️
...e/langflow/agentic/services/helpers/flow_loader.py	90.67% <100.00%> (-1.70%)	⬇️
.../agentic/services/helpers/intent_classification.py	91.13% <100.00%> (-3.74%)	⬇️
... and 53 more

... and 182 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[github-actions]

Frontend Unit Test Coverage Report

Coverage Summary

Lines	Statements	Branches	Functions
	47.14% (67831/143864)	70.42% (9650/13703)	45.77% (1571/3432)

Unit Test Results

Tests	Skipped	Failures	Errors	Time
5448	0 💤	0 ❌	0 🔥	19m 17s ⏱️

[github-actions]

✅ Migration Validation Passed

All migrations follow the Expand-Contract pattern correctly.

[keval718]

We pull inProgressTask out here because it is transient, but flowProposalSnapshot still rides along in ...rest, so it gets persisted. That field is a full cloneDeep of nodes+edges (set in handleApplyFlowProposal), so every applied proposal writes an entire canvas copy into the session localStorage. On a big flow, or after a few proposals in one session, that can balloon and eventually blow past the ~5MB quota — at which point the whole setItem throws and we lose the session save, not just the snapshot.

If keeping revert working across a reload is intentional, can we bound it? e.g. only keep the snapshot while flowProposalStatus === "applied", or keep just the latest one. If it is only meant to power the in-session undo, I think we can drop it in the destructure the same way we do inProgressTask.

[Cristhianzl]

Good catch — you are right. The snapshot rides along in ...rest, and since it is a full cloneDeep of the canvas it can get heavy fast on big flows. And it is worse than just the snapshot: saveSessionsToStorage swallows the quota error, so once setItem starts throwing we silently lose the whole session save.

Cross-reload revert was never the intent here — that is what the restore-point path covers (restoreVersionId + flow versions), and that one survives reloads fine because it lives on the backend. The snapshot only exists to power the in-session undo on the proposal card. So I will go with your second option and drop it in the destructure, same as inProgressTask. After restoring a session the card just renders with canRevert=false and handleRevertFlowProposal already no-ops without a snapshot, so nothing else needs to change.

Will push the fix with a serialization test. Thanks!

[github-actions]

Build successful! ✅
Deploying docs draft.
Deploy successful! View draft

[keval718]

FE looks good to me