Langflow

Knowledge base DB Providers - Langflow 1.10 Choose where your Knowledge Base embeddings are stored. Before Langflow 1.10, Knowledge Bases were primarily tied to local Chroma storage. Now, DB Providers make it easier to configure different vector database backends and create Knowledge Bases using the provider that fits your infrastructure. With the launch of Langflow 1.10, Knowledge Bases support configurable DB Providers: - Chroma Local remains the default provider - Chroma Cloud can be configured as a managed backend - OpenSearch can be enabled and connected as a vector database provider - New Knowledge Bases can be created using the selected DB Provider - Existing Knowledge Bases keep the backend they were created with Example: In this demo, the user opens Settings → DB Providers, reviews the available database providers, and shows that Chroma Local is active by default. Then, OpenSearch is enabled, configured with connection details, tested, and saved as the active provider. After that, the user creates a new Knowledge Base and selects OpenSearch as the backend. Once created, the Knowledge Base keeps that DB Provider, while existing Knowledge Bases remain unchanged. Why it matters: - Gives teams more control over where embeddings are stored - Makes Knowledge Bases more flexible for different infrastructure needs - Supports retrieval workflows beyond local Chroma storage - Keeps the Knowledge Base workflow familiar while adding backend flexibility 👉 Explore Knowledge base DB Providers in Langflow 1.10: https://lnkd.in/dvAsHjJH

Claude Mythos is here, so is Fable, and so are 7 new Microsoft AI models! But what does it take to build with them? Good memory and search, and good evals and goals. Read the latest on all of this in this week's AI++.

Langflow 1.10 - Build complete flows from natural language, directly inside the canvas In Langflow 1.9, Langflow Assistant introduced custom component generation from prompts. With Langflow 1.10, it goes further: creating a full flow is now much easier. You can describe what you want to build, and the assistant helps generate the connected workflow structure for you. Building an agent flow usually means selecting components, wiring them together, configuring models, adding tools, and testing the result manually. With the launch of Langflow 1.10, Langflow Assistant helps turn an idea into a working flow inside the visual builder: - Natural language → complete flow generation - Automatic component selection and wiring - Agent configuration based on the requested use case - Custom component creation when the flow needs new functionality - Canvas updates after approval, keeping the user in control Example: In this demo, the user asks Langflow Assistant to create a restaurant attendant agent flow. The assistant builds the flow with Chat Input, an Agent, Chat Output, and a Calculator tool, configures the model, runs the flow, and validates the response in the Playground. Then, the user asks for an optimization: adding a simple restaurant menu tool with food and drink recommendations. The assistant generates a custom menu component, adds it to the canvas, connects it to the Agent as a tool, and the flow can immediately answer questions like “What do you suggest I drink?” Why it matters: - Makes flow creation faster and easier - Reduces manual setup inside the canvas - Turns natural language prompts into connected workflows - Makes it easier to iterate on agents directly from the canvas Combines flow generation, component creation, tool wiring, and testing in one workflow. 👉 Explore Langflow Assistant in Langflow 1.10: https://lnkd.in/dMcsv6W5

🚀 Langflow 1.10 is live Langflow 1.10 introduces new capabilities for building complete AI workflows with Langflow Assistant, adding persistent memory to flows, managing knowledge bases with flexible database backends, and making Langflow more accessible to global teams. This release expands Langflow Assistant from custom component generation to complete flow building, introduces Memory bases for persistent semantic memory, adds configurable DB Providers for knowledge bases, and brings the Langflow interface to multiple languages. What’s new in this release: 🔹 Langflow Assistant: build complete flows Langflow Assistant can now build entire flows, not only individual components. Describe what you want to build in natural language, and the assistant generates a complete, connected flow ready to run in the canvas. 🔹 Memory bases Memory bases introduce persistent semantic memory for Langflow flows. They store conversation messages in a vector-based memory layer and let flows retrieve relevant past context through semantic search. 🔹 Knowledge base DB Providers Knowledge bases now support configurable database backends through DB Providers, including Chroma, Chroma Cloud, and OpenSearch. This gives users more flexibility to choose where embeddings are stored when creating retrieval workflows. 🔹 Internationalization The Langflow interface is now available in multiple languages, including English, French, Spanish, German, Portuguese, Japanese, and Chinese, making Langflow more accessible for global teams and developers who prefer to work in their native language. Updates in 1.10: 🔹 Redis-backed worker queue Langflow now supports a Redis-backed worker queue for multi-worker deployments, allowing flow build events to be shared across multiple Gunicorn/Uvicorn workers while keeping the default in-memory queue unchanged. 🔹 File System component The new File System component gives agents sandboxed access to files on disk, with an optional Read Only mode to restrict agents to read and search operations only. 🔹 Extension bundles Langflow 1.10 introduces Extension bundles, packaging component providers as standalone pip packages that can be versioned and released independently from the core Langflow package. 🔹 Python 3.14 support Langflow now supports Python 3.10 through 3.14 on macOS, Linux, and Windows. Langflow Docker images now use Python 3.14. 🔹 DB2 component The lfx-ibm bundle now includes an IBM Db2 component, which can be added to flows to connect and query IBM Db2 databases. This makes it easier for enterprise users running Db2 to integrate existing data infrastructure into Langflow workflows. 👉 Explore Langflow 1.10: https://lnkd.in/dMcsv6W5

MCP was adopted amazingly fast by the AI community. Everyone jumped on this bandwagon and for good reason. It's the connective tissue between agents, our app, and the external world. But MCP token usage can get out of hand. The silver lining - there are open source tools like MCPJam that are tailor made for those of us working on MCP services to clean up the bloat and serve better, faster, and richer experiences. Projects like OpenRAG, Langflow, and anyone deploying MCP services can all benefit. Come, listen to my conversation with Prathmesh Patel, CEO of MCPJam, and Mike Fortman, lead developer for OpenRAG, as Prathmesh shows us how MCPJam can help and we discuss some new trends in the MCP space. Pick your poison, then listen to the antidote: YouTube: https://lnkd.in/ehVD-AMy Spotify: https://lnkd.in/eabWxKB7 Apple: https://lnkd.in/ecfbewG9 #MCP #AI #opensource #developer #openrag #langflow #mcpjam

Knowledge Base - Langflow Feature The Knowledge Base feature allows you to create and manage vectorized datasets directly inside Langflow. Instead of configuring a full retrieval flow manually, you can create a knowledge base by defining a name, selecting an embedding model, selecting a supported storage provider, and adding source files. Once created, the source data is chunked, embedded, and stored as a reusable knowledge layer that can be connected to flows. Using Knowledge Bases, you can: - ingest files into a managed knowledge base - configure chunk size, overlap, and separator settings - preview chunks before building the knowledge base - track ingestion status and metadata - reuse the same knowledge base across flows - retrieve relevant results through the Knowledge component The Knowledge component supports both ingestion and retrieval workflows. In ingestion mode, data is added to the selected knowledge base. In retrieval mode, a query is used to search the embedded content and return relevant results to the flow. In the video, a new knowledge base is created from the Knowledge page. A name is defined, an embedding model is selected, and Chroma Local is selected as the storage provider in this example. A source file is uploaded, then the chunking configuration is reviewed before creation. The build step shows a chunk preview, allowing the user to inspect how the content will be split before embedding. After the knowledge base is created, its status changes while the data is ingested. The video then moves to the canvas, where the Knowledge component is added to a flow and connected to the newly created knowledge base. The component is switched from ingest to retrieve mode, a query is executed, and the output is inspected as a table of retrieved results. This shows the full path from source file to searchable context inside a Langflow workflow. Available in Langflow: https://lnkd.in/dZ5C5Sav

Langflow Use Case: Document Data Intelligence Transform contracts and legal documents into structured, machine-readable data. This Langflow workflow processes unstructured documents, extracts key information such as company details, legal identifiers, contact information, and representative data, and validates the results against a predefined schema. The output is formatted as structured data that can be integrated directly into databases, CRMs, ERPs, and other business systems. Use it to automate contract processing, reduce manual data entry, and standardize information extraction across large document volumes. 🔗 Template: https://lnkd.in/d_SSEnXy

What happens when an AI agent is allowed to take action? In agentic workflows, the model is not only generating answers. It may call tools, trigger workflows, update systems, apply rules, or make decisions that affect real business processes. That is where prompts alone can become fragile. Our latest blog explores how Langflow Policies helps address this by turning natural-language rules into guarded tools. Instead of treating a business rule as extra prompt text and hoping the model follows it, Policies attaches the rule to the tool itself. The agent can still reason and plan normally, but the tool call only executes if it passes the policy check. In the blog, we walk through a practical clinic appointment example: Only Gold customers are eligible for a 10% discount on all doctor appointments. From there, we show how Langflow uses ToolGuard to generate guard logic, connect it to an agent, and enforce the policy at runtime. The post also includes a step-by-step video demo showing the full workflow inside Langflow. If you are building agents that interact with real tools, this is a useful pattern to understand. Prompts guide behavior. Policies constrain execution. Read the full blog and watch the demo: https://lnkd.in/dttkaSkQ

2026 is truly the year we harness agents. In order to do that we need to learn how to harness them. In this edition of AI++ we deep dive into what a harness is, how you build one, and how Langflow helps you control your agents. We also take a look at what happened at Google I/O and where the MCP project is going next. There's a new version of the protocol imminent, so check in to see what's going to change.