Langflow is a new, visual way to build, iterate and deploy AI apps.

Industry

Software Development

Company size

11-50 employees

Headquarters

Uberlândia, Minas Gerais

Type

Self-Owned

Founded

2020

Specialties

AI, Generative AI, GenAI, RAG, and Machine Learning

• 16,858 followers

  1w

  When one vector is not enough for RAG Single-vector retrieval works well for many workflows. But when documents get longer, denser, or visually complex, compressing everything into one embedding can lose important context. A long technical report, a PDF full of tables, a slide deck, or a scanned document often contains meaning across tokens, layout, charts, and visual structure. That is where multi-vector retrieval changes the game. In our latest blog, we break down how Langflow 1.11 introduces first-class multi-vector retrieval with the new lfx-nextplaid extension bundle. With NextPlaid, Langflow flows can now use ColBERT-style late interaction and ColPali-style visual document retrieval without custom glue code. What this enables: - Token-level retrieval instead of relying on a single pooled document vector - Better search quality for long and information-dense documents - Visual retrieval for PDFs, tables, charts, slide decks, and scanned pages - PLAID indexing for faster, more memory-efficient multi-vector retrieval - New NextPlaid and vLLM Multivector Embeddings components inside Langflow The benchmark results are significant. On long technical reports, text multi-vector retrieval via PLAID reached 94.7 Recall@10, compared to 73.7 with single-vector retrieval. For image retrieval, multi-vector retrieval reached 89.7, compared to 22.6 with a single-vector baseline. If you build RAG workflows over complex documents, technical reports, enterprise knowledge, or visual PDFs, this deep dive is worth reading. 👉 Read the full blog: https://lnkd.in/d98-newT

• 16,858 followers

  2w

  🚀 Langflow Desktop 1.11 is now available Langflow 1.11 brings new capabilities for building agentic workflows with more control, interoperability, and API-driven execution. This release includes: - Human-in-the-Loop for review and approval gates - A2A protocol support for agent-to-agent interoperability - AG-UI-compatible streaming through the Workflow API If you are already using Langflow Desktop, upgrade in the app to version 1.11. To install Langflow Desktop for the first time, visit: https://lnkd.in/djRBgVNa

• 16,858 followers

  3w

  🔹 AG-UI streaming for the Workflow API - Langflow 1.11 Bring Langflow workflows into real-time agent interfaces. With the launch of Langflow 1.11, the v2 Workflow API now supports AG-UI-compatible streaming through POST /api/v2/workflows. Developers can run a workflow in stream mode, set stream_protocol: "agui", and receive Server-Sent Events in a format designed for AG-UI clients. Instead of waiting for the full response at the end of a run, external applications can follow the workflow as it executes: - Assistant messages stream as they are generated - Tool activity is surfaced during the run - Workflow steps and state updates are emitted - Completion and error events are sent back to the client The default Langflow EventManager protocol remains available, while AG-UI adds a standard event format for teams building custom agent experiences outside the Langflow canvas. Why it matters: - Makes Langflow easier to integrate into external applications - Enables real-time agent interfaces powered by Langflow workflows - Gives AG-UI clients a compatible streaming format - Helps developers build custom copilots, dashboards, and agent experiences beyond the canvas 👉 Explore AG-UI streaming in Langflow 1.11: https://lnkd.in/dTHiTK9D

• 16,858 followers

  3w

  A2A protocol support - Langflow 1.11 Connect Langflow flows to broader agent ecosystems. As agentic systems become more distributed, agents need a standard way to discover, call, and collaborate with each other. With the launch of Langflow 1.11, Langflow now supports the Agent2Agent (A2A) protocol, making it possible to publish Langflow flows as callable agents and connect them with other A2A-compatible systems. With A2A support, you can: - Publish Langflow flows as agents that other systems can call - Expose flows through agent discovery and A2A-compatible endpoints - Call remote A2A agents from inside a Langflow flow - Build more modular multi-agent workflows across different environments Example: A Langflow flow can be published as an A2A agent, making it discoverable and callable by external A2A clients. Inside Langflow, you can also use the A2A Agent component to call another published Langflow agent or connect to a remote A2A-compatible agent. This means Langflow can both expose agents to the outside world and call agents from other systems as part of a larger workflow. Why it matters: - Turns Langflow flows into interoperable agents - Makes it easier to connect agentic systems - Supports more flexible multi-agent architectures - Helps developers build workflows where agents collaborate across tools, platforms, and environments 👉 Explore A2A support in Langflow 1.11: https://lnkd.in/dTHiTK9D

• 16,858 followers

  4w

  🔹 Human-in-the-Loop - Langflow 1.11 Keep humans in control when it matters most. As agentic workflows become more capable, some decisions should not happen automatically. Certain steps may need human review, approval, or a manual choice before the flow continues. With the launch of Langflow 1.11, Human-in-the-Loop lets flows pause execution, create a checkpoint, and wait for a human decision before moving forward. Human-in-the-Loop supports: - Flow pauses at the point where human input is required - Stateful checkpoints before execution continues - Approve, Reject, or custom user choices - Branch-based routing with the Human Input component - Tool approval gates before an agent uses selected tools - Resuming from the checkpoint without rerunning previous steps Example: In this video, the flow receives a user request to search for how many players are currently online playing a game. Before the agent continues, the Human Input component pauses the flow and asks for a human decision: should the search be about Tibia or World of Warcraft? Once the user selects one of the available options, the flow resumes from that checkpoint, follows the selected branch, runs the corresponding web search, and returns the result. This shows how Human-in-the-Loop can turn a fully automated workflow into a controlled decision process, where the agent continues only after the right human input is provided. Why it matters: - Adds control to agentic workflows - Keeps sensitive or important actions under human review - Makes approval and decision flows easier to build visually - Allows workflows to branch based on human input - Keeps agents autonomous while preserving human oversight 👉 Explore Human-in-the-Loop in Langflow 1.11: https://lnkd.in/dTHiTK9D

• 16,858 followers

  1mo Edited

  🚀 Langflow OSS 1.11 is live Langflow 1.11 introduces new capabilities for building agentic workflows that are more controlled, more interoperable, and easier to run through external applications. This release adds Human-in-the-Loop for human review and approval gates, A2A protocol support for agent-to-agent interoperability, and AG-UI-compatible streaming through the Workflow API. What’s new in this release: 🔹 Human-in-the-Loop Langflow flows can now pause execution, create a stateful checkpoint, and wait for a human decision before continuing. This means developers can add human review into critical steps of a workflow, including approval gates before an agent uses specific tools. After the reviewer approves, rejects, or selects a configured action, the flow can resume from the checkpoint and continue through the selected path, without rerunning previously completed steps. This makes it easier to build workflows where agents can act autonomously, while sensitive or high-impact actions remain under human control. 🔹 A2A protocol support Langflow now supports the Agent2Agent protocol. You can publish a Langflow flow so other agents can call it, and you can also call remote A2A agents from inside a Langflow flow. This gives developers a clearer path to connect Langflow agents with external agent systems, published flows, and broader multi-agent ecosystems. With A2A support, Langflow flows can become callable agents, and Langflow agents can interact with other A2A-compatible agents as part of a larger workflow. 🔹 Workflow API with AG-UI streaming Langflow 1.11 expands the Developer v2 Workflow API with AG-UI-compatible streaming. The /api/v2/workflows endpoint can run workflows programmatically and stream events using stream_protocol: "agui". This gives developers a more flexible foundation for building external applications, interfaces, and agent experiences powered by Langflow workflows. The default Langflow EventManager protocol remains available, while AG-UI adds a standard event format for teams building richer external agent interfaces. Also included in 1.11: - Unified Data Operations component (consolidating Text, JSON, and Table Operations) - New provider component bundles (NextPlaid, Paddle, Oracle, and Valkey) - LFX engine-only install option (install the executor without bundle components) 👉 Explore more about Langflow 1.11: https://lnkd.in/dTHiTK9D

• 16,858 followers

  1mo

  Smart Router - Langflow Component Route by meaning, not by regex. The Smart Router component is an LLM-powered alternative to traditional If/Else routing inside Langflow flows. Instead of matching strings or writing rigid conditions, Smart Router uses a connected language model to classify incoming input and send it to the matching output port. Routes are defined through a table, where each row can include: - Route Name: the output path created by the component - Route Description: guidance for how the LLM should classify the input - Route Message: an optional custom message returned for that route For example, a support flow can route incoming messages into Positive, Negative, or Neutral paths based on meaning, urgency, and tone (not just keywords). When the input is classified, the component sends it through the corresponding route. For unmatched cases, an Else output can be enabled. Additional Instructions can also be used to provide domain-specific context to the classification step. This is useful when a flow needs flexible decision logic without hard-coding every possible input variation. Available in Langflow: https://lnkd.in/du-wr9hF

• 16,858 followers

  1mo

  File Processing - Langflow Components Agents often need to work with large collections of files, but loading every file directly into the prompt does not scale well. The File Processing component helps separate file discovery from file retrieval inside Langflow workflows. Two components work together in this pattern: - File Description Generator creates searchable natural-language descriptions for each file - File Content Retriever exposes tools that allow an agent to retrieve specific file content by path This gives the agent a more controlled way to work with files. First, file descriptions can be generated and indexed so the agent can identify which files are relevant to a task. Then, when the actual content is needed, File Content Retriever lets the agent request a specific file and return it as: File Content: file content as text Table: file content as a table for formats like CSV, Excel, Parquet, JSON, and TSV This is useful when an agent needs to inspect specific documents, work with tabular data, or retrieve content from a large file set without pushing everything into the context window upfront. File maps and generated descriptions can also be cached, reducing repeated processing across runs. Available in Langflow: https://lnkd.in/dFYuABbG

• 16,858 followers

  1mo

  Message History - Langflow Component The Message History component stores and retrieves chat messages inside a Langflow flow. It can use Langflow storage by default or connect to an external memory provider when memory needs to live outside the built-in database. The component works in two modes: Retrieve: fetches stored messages and passes them into the flow Store: saves new messages after a flow runs This makes memory explicit in the workflow instead of keeping it hidden inside the chat interface. Using Message History, you can: - retrieve past messages before calling an LLM - store model responses after execution - separate memory by session ID - use external memory providers when needed - return memories as messages or tabular data - build flows that analyze or reuse previous conversation context A typical pattern: - Retrieve memory - Pass it into a prompt or model - Generate a response - Store the new message In the creative, the component is shown in Store mode. The inspection panel exposes fields such as External Memory, Sender, Sender Name, Session ID, and Context ID, making it possible to control where messages are stored and how they are grouped. This is useful when a flow needs memory that can be inspected, filtered, reused, or connected to external storage. Available in Langflow: https://lnkd.in/dwv7__tB