← All reportsPublic report

langflow-ai/langflow

Langflow is a powerful tool for building and deploying AI-powered agents and workflows.

154k stars Python View on GitHubprofiled 11d ago
Try GitZoid
381PRs this week
24Contributors
2Deps scanned
0Issues found
01 · Repo overview

How langflow is put together

Langflow is a visual, low-code IDE for building LangChain-based LLM/agent workflows, packaged as a Python web application with a bundled React frontend. The backend is a FastAPI/Starlette app (entry `langflow.main:create_app`) served via uvicorn/gunicorn, using a uv workspace that splits core (`langflow-base`), the `lfx` execution library, an SDK, and many provider bundles (`lfx-*` for OpenAI, Anthropic, Azure, Google, IBM, Amazon, Ollama, etc.). Users design flows in a drag-and-drop React Flow canvas that persist to a SQL database (SQLite by default, with Postgres/pgvector/opensearch options) and execute component graphs on the server. Data flows from the React SPA over an HTTP API to the FastAPI backend, which runs graph components that call external model/tool providers.

Languages

PythonTypeScriptJavaScript

Frameworks

FastAPI/Starlette (uvicorn/gunicorn)LangChainReact 19React Flow (@xyflow/react)ViteTailwindCSSZustandTanStack QueryStorybookPlaywrightJest

Datastores

SQLite (default)PostgreSQLpgvectorOpenSearchRedis/fakeredis (testing)ClickHouse (optional)Couchbase (optional)Oracle (bundle)

Infrastructure

Docker / docker-composeNginx (frontend serving)GitHub Codespaces / devcontainerRender (render.yaml)Prometheus/Grafana-Loki observabilityQEMU microVM sandbox (opt-in)

Major components

Backend core (langflow-base / langflow.main)

FastAPI application factory, API routes, auth, and orchestration of flow execution and persistence.

lfx execution library (src/lfx)

Core graph/component execution engine and pluggable services including an MCP integration layer.

Provider bundles (src/bundles/*)

Independently versioned lfx-* packages integrating external model/tool/data providers (OpenAI, Anthropic, Azure, Google, IBM, Amazon, Ollama, Oracle, etc.).

React frontend (src/frontend)

Vite/React SPA providing the drag-and-drop flow editor built on React Flow, state via Zustand and TanStack Query.

SDK (src/sdk)

Python SDK package for programmatic interaction with Langflow environments.

langflow-stepflow (src/langflow-stepflow)

Separate workspace package for stepflow functionality.

Deployment & infra (deploy/, docker/, render.yaml)

Container images, compose stacks, Nginx config, and observability setup for running Langflow.

Agent skills & tooling (.agents/skills)

Repo-specific code-review, testing, and accessibility skill definitions used in development workflows.

Over the past several weeks Langflow shifted from a burst of major feature work — unified knowledge/memory bases on shared vector backends, provider governance, SSO foundations, and OpenTelemetry-based observability — into a sustained hardening phase focused on security, access control, and release stabilization across the 1.11.x and 1.12 lines. Recent weeks are dominated by security backports (SSRF, sandboxing, share-permission fixes), accessibility improvements, and CI/test reliability, with feature velocity slowing as the team locks down and stabilizes shipped work.

Week by week

2026-08-24A quiet release-stabilization week folding fixes into the 1.11.5 line.latest3 changes

Fix

Handle tagged Ollama embedding models

Model names like 'bge-m3:latest' now correctly resolve to the Ollama provider instead of defaulting to OpenAI.

Chore

Stabilize nightly Windows browser tests

Made the automated nightly Windows test runs more reliable so flaky failures stop blocking releases.

Chore

Merge 1.11.5 release into main

Folded the latest patch-release changes back into the main development branch.

2026-08-17Security backports and release plumbing dominated as fixes were carried across the 1.11.4, 1.11.5, and 1.12 branches.6 changes

Fix

Backport several security hardening fixes

Carried MCP execution, generated-code boundary, and other security fixes into the 1.11.5 release line.

Fix

Close model-provider SSRF gaps

Tightened protections against requests being tricked into reaching unintended internal servers.

Feature

Opt-in bypass for multi-worker queue guard

Added an advanced, headless-only setting to run multiple workers without the shared Redis queue.

Fix

Keep the persisted flow on header refresh

Prevented the open flow from being lost when the header's flow list refreshes.

Docs

Record security triage notes

Documented why four repeatedly-flagged scanner findings can't be fixed by a version bump.

Fix

Keep skipped package releases green in CI

Stopped the release automation from failing when a package release is intentionally skipped.

2026-08-10A heavy week combining broad access-control and security fixes, accessibility work, and new observability tracing across the 1.11.4 and 1.12 lines.6 changes

Fix

Restrict and repair public flow sharing

Limited public shares to execute-only and made public playground links work correctly for anonymous visitors.

Feature

Project upsert API endpoint

Added an API endpoint to create-or-update projects by a stable id.

Fix

Accessibility fixes across the canvas

Improved keyboard focus visibility, field labeling, and menu navigation for screen-reader and keyboard users.

Feature

Distributed tracing for outbound calls

Added tracing spans for agent-to-agent calls so runs can be observed end to end.

Fix

Rebuild drifted built-in components

Automatically repairs out-of-date built-in components instead of refusing to open the flow.

Fix

Withhold log bodies from telemetry export

Stopped log message contents from being sent to external observability backends by default.

2026-08-03A large security-and-hardening push landed, closing SSRF, file-handling, and access-control gaps while stabilizing tests.6 changes

Feature

Opt-in sandboxed code execution

Added a secure micro-VM backend option to isolate components that run user code.

Fix

Harden file and URL handling

Contained file references, made cloud file cleanup storage-aware, and rejected ambiguous or unsafe URLs.

Fix

Enforce admin-only component builds

Restricted who can build components and tightened approval of the allowed component catalog.

Feature

Shared database-backed policy bundle

Introduced a common policy store to govern provider and catalog access across the app.

Fix

More reliable flow updates and requests

Retried flow saves on database locks and ended request transactions before long-running work.

Fix

Sanitize markdown rendering

Closed gaps in the markdown sanitizer to prevent unsafe content from rendering.

2026-07-27The busiest week delivered major foundations — unified knowledge/memory bases, provider governance, and full OTLP observability — alongside heavy security and frontend cleanup.6 changes

Feature

Unified knowledge and memory bases

Consolidated knowledge bases and memory bases onto a common vector-database backend with pgVector support.

Feature

Provider governance and approved-provider policy

Established a foundation to govern which model providers are approved and how roles are granted.

Feature

Export telemetry to any OTLP backend

Made application metrics and traces exportable to standard observability tools, including event-loop and DB-pool health metrics.

Fix

Harden public flow and API execution

Guarded public flow execution, sanitized public workflow calls, and scoped public projects to their owners.

Refactor

Decompose large frontend pages

Broke up big provider, dropdown, and prompt components into smaller reusable hooks and helpers.

Fix

Steadier model selection and flow navigation

Replaced models of disconnected providers and stopped stale flow-list refreshes from bouncing users around.

03 · Security check

Dependencies and code review

2 dependencies scanned

Dependency advisories

Security Watch

No known advisories across 2 scanned dependencies.

No known advisories in the scanned dependencies.

Code review

No concrete code-level issues confirmed.

Get this report every week for your repos.

GitZoid learns each repo, reports what changed, and flags what needs attention. One flat price for the whole team.

$19 a month, flat · First 10 outputs free · No card required