docs/docs/Support/release-notes.mdx
import Icon from "@site/src/components/icon"; import McpIcon from '@site/static/logos/mcp-icon.svg';
This page summarizes significant changes to Langflow in each release. For all changes, see the Changelog.
Due to strict SemVer requirements, Langflow Desktop can have different patch versions than the core Langflow OSS Python package, but the major and minor versions are aligned.
:::warning Whenever possible, the Langflow team recommends installing new Langflow versions in a new virtual environment or VM before upgrading your primary installation. This allows you to import flows from your existing installation and test them in the new version without disrupting your existing installation. In the event of breaking changes or bugs, your existing installation is preserved in a stable state. :::
To avoid the impact of potential breaking changes and test new versions, the Langflow team recommends the following upgrade process:
Recommended: Export your projects to create backups of your flows:
curl -X GET \
"$LANGFLOW_SERVER_URL/api/v1/projects/download/$PROJECT_ID" \
-H "accept: application/json" \
-H "x-api-key: $LANGFLOW_API_KEY"
To export flows from the visual editor, see Import and export flows.
Install the new version:
Import your flows to test them in the new version, upgrading components as needed.
When upgrading components, you can use the Create backup flow before updating option if you didn't previously export your flows.
If you installed the new version in isolation, upgrade your primary installation after testing the new version.
If you made changes to your flows in the isolated installation, you might want to export and import those flows back to your upgraded primary installation so you don't have to repeat the component upgrade process.
Langflow 1.12 consolidates the runnable application on langflow-base.
pip install langflow-base now provides the UI, API, platform services, built-in
components, and the langflow command without provider extension
distributions. pip install langflow installs that same base version and adds
the curated standalone lfx-* extensions.
langflow-base version alignment
The runnable package starts at langflow-base==1.12.0 and remains
version-aligned with langflow through stable, release-candidate, patch,
and nightly releases. A dependency such as langflow-base~=0.12 remains on
the old package line and must be updated explicitly to select 1.12.
Before upgrading an existing environment, remove the retired package with
pip uninstall langflow-core; leaving an older langflow-core installed
can create conflicting dependency requirements.
Base command and image tags
The base package provides the langflow executable, not a
langflow-base executable. The canonical bundle-free container is
langflowai/langflow:base-VERSION; normal langflowai/langflow:VERSION
tags contain the curated distribution. The previous core-* image profile
is no longer published.
Cassandra and Policies components are extensions
Cassandra components now ship in lfx-datastax, and the
PoliciesComponent ships in lfx-toolguard. The full langflow package
includes both. Existing component class names and legacy Python import
paths remain compatible. For a base-only environment, add the extensions
with pip install "lfx[cassandra,toolguard]".
Policies guard caches are now isolated under
tmp_toolguard/{user}/{flow}/{component}/{project}. Existing project-only
caches are not reused after upgrading; run Generate once before using
Guard in an upgraded flow.
The default langflow provider set is smaller
arXiv, DuckDuckGo, EmpirioLabs, Exa, Firecrawl, NextPlaid, Paddle, and
Valkey are no longer installed by plain pip install langflow. Use
pip install "langflow[bundles]" to add those standalone extensions
together with every no-Torch long-tail bundle.
PyTorch remains opt-in
The default base and curated Langflow installations do not install PyTorch or TorchVision. Components that require them must be installed through an explicit opt-in profile.
IBM watsonx Orchestrate clients move to the IBM extension
The base deployment adapter remains feature-flagged, but its client
dependencies now ship with lfx-ibm. Install lfx-ibm or use the
compatibility extra langflow-base[ibm-watsonx-clients] before enabling
the adapter in a base-only environment.
LiteLLM remains optional in base-only environments
langflow-base no longer imports LiteLLM by default. Install
langflow-base[litellm] when custom components or IDE tooling depend on
the LiteLLM compatibility surface.
Highlights of this release include the following changes. For all changes, see the Changelog.
:::tip
If Langflow fails to start with a get_body_field error after installing version 1.11.0, see get_body_field error after installing Langflow.
:::
Short LANGFLOW_SECRET_KEY upgrade compatibility
Langflow 1.10.1 changed how secrets shorter than 32 characters derive the Fernet key used for encrypted credentials. Langflow 1.11.x retains read compatibility with credentials encrypted by the earlier derivation while continuing to use the current SHA-256 derivation for all new writes.
If your deployment uses a LANGFLOW_SECRET_KEY shorter than 32 characters, keep it unchanged during the initial upgrade so Langflow can continue to decrypt existing credentials.
Langflow logs a startup warning for this configuration because short secrets aren't recommended for production.
To replace the key, first record or export the stored credentials, set a randomly generated key of at least 32 characters, and then re-enter the credentials; changing the key directly invalidates ciphertext created with the old key.
Default superuser password removed
Langflow no longer creates or accepts the legacy langflow/langflow default superuser credentials.
If LANGFLOW_AUTO_LOGIN=False, set LANGFLOW_SUPERUSER_PASSWORD to a strong password before startup.
The legacy value langflow is not allowed, even if LANGFLOW_AUTO_LOGIN=True.
export LANGFLOW_SUPERUSER_PASSWORD=SUPERUSER_PASSWORD
Replace SUPERUSER_PASSWORD with a strong password for the Langflow superuser.
If LANGFLOW_AUTO_LOGIN=true, setting LANGFLOW_SUPERUSER_PASSWORD is optional. If you omit it, Langflow generates a random bootstrap password for the auto-login account.
If you're running Langflow with Docker and LANGFLOW_AUTO_LOGIN=false, pass the password at startup:
docker run -d \
--name ${CONTAINER_NAME} \
--restart unless-stopped \
-p 7860:7860 \
-e LANGFLOW_HOST=0.0.0.0 \
-e LANGFLOW_PORT=7860 \
-e LANGFLOW_AUTO_LOGIN=false \
-e LANGFLOW_SUPERUSER_PASSWORD=SUPERUSER_PASSWORD \
-v langflow-data:/app/langflow \
${IMAGE}
Replace SUPERUSER_PASSWORD with a strong password for the Langflow superuser.
Docker images disable auto-login by default
Official Langflow Docker images set LANGFLOW_AUTO_LOGIN=false.
You must set LANGFLOW_SUPERUSER_PASSWORD (and optionally LANGFLOW_SUPERUSER) before the container can start, unless you explicitly set LANGFLOW_AUTO_LOGIN=true.
For more information, see Docker image defaults.
Docker runtime home moved to /app/data
Patched 1.11 Docker images set HOME=/app/data so the non-root runtime user has a writable home directory.
When LANGFLOW_CONFIG_DIR isn't set, the default config directory therefore changes from /opt/app-root/src/.cache/langflow to /app/data/.cache/langflow; SQLite follows that directory when LANGFLOW_SAVE_DB_IN_CONFIG_DIR=true.
Persist /app/data or set LANGFLOW_CONFIG_DIR to your existing writable data directory before upgrading to preserve existing data.
Long-tail component bundles are now opt-in
Starting in Langflow 1.11.x, pip install langflow no longer installs the long-tail providers in the lfx-bundles metapackage.
If an existing flow uses components such as Chroma, Ollama, Qdrant, Redis, PGVector, Tavily, Wikipedia, MongoDB, Weaviate, or Milvus, install the metapackage in the same environment after upgrading:
pip install lfx-bundles
You can instead install one provider and its dependencies, for example pip install "lfx-bundles[qdrant]".
Missing-provider errors use the same install guidance.
Graduated providers such as OpenAI, Anthropic, Cohere, and Exa remain part of the default Langflow installation through their standalone lfx-* packages.
For the complete package list and provider-specific commands, see Extension bundle list.
PyTorch components are opt-in by default
CUGA, Code Agents, and Docling local conversion are excluded from the default uv pip install langflow installation because they require PyTorch.
To install these components, see Torch opt-in installs.
Workflow API request schema (Beta)
The v2 Workflow API (Beta) request body and interaction pattern changes in 1.11.
Any client that calls POST /api/v2/workflows against 1.10.x must update as follows:
Example 1.10.x request:
{
"flow_id": "67ccd2be-17f0-8190-81ff-3bb2cf6508e6",
"background": false,
"inputs": {
"ChatInput-abc.input_value": "what is 2+2",
"LLMComponent-123.temperature": 0.7
}
}
Example 1.11.x request:
{
"flow_id": "67ccd2be-17f0-8190-81ff-3bb2cf6508e6",
"input_value": "what is 2+2",
"mode": "stream",
"stream_protocol": "agui",
"tweaks": {
"LLMComponent-123": {
"temperature": 0.7
}
}
}
For more information, see Workflow API (Beta).
Input Schema pane replaced by in-component Parameters
The Input Schema pane under Share > API access is removed, but the fields can still be exposed to requests.
The tweaks object in API requests is unchanged.
Endpoint Name remains available from the API access pane.
For details, see Tweaks (API inputs).
Human-in-the-Loop (HITL)
Human-in-the-Loop (HITL) pauses an agent when the agent calls a tool and creates a stateful checkpoint. After a human responds by approving, rejecting, or editing the request, the agent resumes from the checkpoint.
For more information, see Human-in-the-Loop.
Agent2Agent (A2A) protocol support
Publish a flow as an A2A agent so other agents can discover and call it, and call remote A2A agents from inside a flow with the A2A Agent component.
For more information, see Use Langflow as an A2A server and the A2A Agent component.
AG-UI compatible streaming for the Workflow API
The v2 Workflow API now supports streaming with the AG-UI (Agent–User Interaction) protocol streaming format.
For more information, see Workflow API (Beta).
OpenAI Compatible model provider
Point Langflow's model provider at any OpenAI-compatible endpoint, and Langflow discovers models live from the /v1/models endpoint.
These models can power flows, Langflow Assistant, and any component that uses Langflow’s global model providers.
For more information, see OpenAI Compatible.
Unified Data Operations component
Text Operations, JSON Operations, and Table Operations are consolidated into a single Data Operations component. Saved flows that use the separate components continue to work. For more information, see the Data Operations component.
IBM watsonx Orchestrate: Python 3.14 compatibility
The ibm-watsonx-orchestrate-core and ibm-watsonx-orchestrate-clients packages are upgraded to version 2.12, which supports Python 3.14.
IBM watsonx Orchestrate is no longer excluded from Python 3.14 installs.
For more information, see Deploy flows on watsonx Orchestrate.
The following optional integrations remain excluded from installations on Python 3.14:
NextPlaid multi-vector bundle
The NextPlaid bundle adds two new components for ColBERT-style multi-vector retrieval.
The NextPlaid vector store, backed by a running NextPlaid server, stores each document as a matrix of token embeddings, and the vLLM Multivector Embeddings component generates the token-level multi-vector embeddings required by NextPlaid.
For more information, see NextPlaid bundle.
PaddleOCR bundle
The Paddle bundle (lfx-paddle) adds a PaddleOCR component that calls the PaddleOCR AI Studio Job API for layout-aware document parsing into Markdown.
For more information, see Paddle bundle.
Oracle Extension bundle
The Oracle bundle adds Oracle Database integration for vector search, document loading, and embeddings.
For more information, see Oracle bundle.
Valkey bundle
The Valkey bundle adds a vector store and chat memory components for Valkey, an open-source Redis fork.
For more information, see Valkey bundle.
LFX is now engine-only
uv pip install lfx now installs the LFX executor only, with no bundle components included.
uv pip install langflow installs the server and curated standalone providers, but not the long-tail lfx-bundles metapackage.
If your flows use long-tail bundle components, install the required packages in the same environment.
About 70 long-tail providers ship in the lfx-bundles metapackage.
Install one provider with uv pip install "lfx-bundles[<bundle-name>]", or install every long-tail provider with uv pip install "lfx[bundles]".
Graduated providers, such as OpenAI, Anthropic, Cohere, and Exa, ship as individual packages and are installed separately with uv pip install lfx-<provider>.
The lfx-bundles long tail is opt-in for both lfx and langflow installations.
For more information, see Extensions overview.
For 1.10.x release notes, see the 1.10.x documentation.
For 1.9.x release notes, see the 1.9.x documentation.
For 1.8.x release notes, see the 1.8.x documentation.
See the Changelog.