# Langflow Visual builder for assembling AI agents and workflows by drag and drop and serving them as an API or MCP server. Canonical profile: https://ki-agentenkompass.de/en/agents/langflow/ Editor: [Michael Meierhoff](https://ki-agentenkompass.de/en/about/) Sources reviewed: 2026-11-05 Basis: provider documentation. Examples are editorial suggestions, not hands-on product tests. ## Overview Langflow is an open-source project. A blog post dated 25 February 2025 names DataStax as parent company and announces IBM’s planned acquisition of DataStax. DataStax now appears under IBM’s name, but the reviewed provider pages do not state completion. - Provider: Langflow - Product family: Langflow - Profile type: Builder for agents and workflows - Environment: Langflow Desktop (macOS, Windows), Python package, Docker image or source; used in the browser - Connections: Model providers, MCP server and client, webhooks, REST API, custom Python components - Setup: Install the desktop app, the uv package or Docker, add a model provider API key and build a flow from a template such as Simple Agent. - Pricing model: Open source under the MIT licence. Models need your own provider access. No hosted offering could be found. ## Key features ### Visual flow editor Components are connected by drag and drop. Python lets you add custom components. Source: https://www.langflow.org ### Agents with tools According to the documentation, the Agent component can use any component as a tool, including other agents and MCP servers. You enable Tool Mode on the component for this. Source: https://docs.langflow.org/agents ### Serve as API and MCP server Flows can be served as an API. Each project gets its own MCP server whose tools MCP clients can use. A flow needs a Chat Output component for this. Source: https://docs.langflow.org/mcp-server ### Use MCP tools The MCP Tools component connects external MCP servers via STDIO, HTTP/SSE or a JSON configuration. Source: https://docs.langflow.org/mcp-client ### Webhook triggers A POST to the webhook endpoint starts a flow. The response only confirms the start, not the result. A Langflow API key is required by default. Source: https://docs.langflow.org/webhook ### Built-in chat memory The Agent component has chat memory enabled by default, grouped by session ID. The default is a SQLite database; another database can be configured. Source: https://docs.langflow.org/memory ### Flow version history You save snapshots of a flow as versions, preview them read-only and restore them. Versions are stored in the installation’s database. Source: https://docs.langflow.org/concepts-flows ## Example workflows ### Serve an agent with tools as an API - Your input: A flow from the Simple Agent template with a configured model provider. - Possible result: The agent answers requests in the Playground and can be called as an API. - What to check: Test in the Playground, read the message logs and keep an eye on model costs. ### Start a flow by webhook - Your input: A flow with a Webhook component that receives JSON from an external application. - Possible result: Each POST starts a run; a Parser component turns the data into text. - What to check: Check results inside the flow, because the response only confirms the start. ## Requirements and limits ### Model access required Requirement: The quickstart uses an OpenAI key as its example. Only one key can be added per provider, and the account needs sufficient credits. Source: https://docs.langflow.org/get-started-quickstart ### Python and uv for the package install Requirement: The Python package needs Python 3.10 to 3.14 and uv. The minimum is a dual-core CPU and 2 GB RAM; more cores and at least 4 GB are recommended. Source: https://docs.langflow.org/get-started-installation ### Desktop without Shareable Playground and Voice Mode Limitation: According to the installation page, some features are unavailable in Langflow Desktop, such as the Shareable Playground and Voice Mode. Source: https://docs.langflow.org/get-started-installation ### Role-based access only via plugin Limitation: Role-based access control needs a registered authorisation plugin, which Langflow does not include. Without one, every authenticated user is allowed. Source: https://docs.langflow.org/authentication-overview ### MCP server without authentication is public Limitation: If a project has no authentication set, anyone can use its MCP server without credentials. The documentation advises this only in trusted environments. Source: https://docs.langflow.org/mcp-server ### Webhook returns no result Limitation: The webhook response only shows that the flow started. The webhook page describes no schedules. Source: https://docs.langflow.org/webhook ### Hosted offering removed Limitation: According to DataStax, hosted Langflow was removed from Astra on 9 April 2026. The Langflow homepage still mentions a free cloud account, but no sign-up for it or other hosted offering could be found. Source: https://docs.datastax.com/en/langflow/index.html ## A good fit if … - you want to build agents and workflows visually and serve them as an API or MCP server - you want to run the software yourself and choose model providers freely - you want to add your own Python components and external MCP tools to a flow ## Less suitable if … - you expect fixed roles and permissions without an extra plugin - you need schedules or approval before tool calls built in - you are looking for a solution hosted by the provider Editorial assessment based on provider documentation. Not a product test. ## Usable AI models Open choice You add your own key per provider in the settings and pick the model in the agent. Included by default are Anthropic, OpenAI, Google, IBM, Cohere, Ollama and an OpenAI-compatible endpoint, among others. Mistral, DeepSeek, xAI, OpenRouter and LiteLLM come through add-on packages. - Model families named: Anthropic (Claude), OpenAI (GPT), Google (Gemini), IBM (Granite), Cohere, Mistral, DeepSeek, xAI (Grok) · Other models can be connected freely - Model choice: You choose - Your own API key: Yes - Router or custom endpoint: Yes - Local models: Yes Documented · Sources accessed on: 2026-11-05 - [Langflow: Use agents](https://docs.langflow.org/agents) - [Langflow: Language Model](https://docs.langflow.org/components-models) - [Langflow: Bundle list](https://docs.langflow.org/extensions-bundle-list) ## Europe & regional availability Website, documentation and telemetry page reviewed. They name no country release or data regions. Reviewed scope: Self-operated software; model providers separately Reviewed on: 2026-11-05 ### Europe: no provider statement Availability still unclear When self-hosting, flows and chat history are stored in a local SQLite database by default. The reviewed pages say nothing about Europe or regions of the cloud offering. On telemetry, the documentation names DO_NOT_TRACK=True to opt out; it does not state the default explicitly. It lists items such as durations and version data, plus the registered email address in the desktop version. Region covered: EU/EEA, United Kingdom and Switzerland Documented reason: The reviewed sources name no country release for this. Official source: https://docs.langflow.org/memory Official source: https://docs.langflow.org/contributing-telemetry Official source: https://www.langflow.org ## Accounts & permissions Based on provider documentation. Permissions depend on the account, integration and organization policies. Review the permissions actually requested before use. Configuration covered: Self-operated Langflow installation with your own model access Sources reviewed: 2026-11-05 ### Account used Documented: According to the documentation, built-in sign-in with username and password is always available and the default; it issues short-lived session tokens. Alternatively an identity provider handles sign-in through a proxy. You add the model provider key in settings under Model Providers. Evidence: https://docs.langflow.org/authentication-overview Evidence: https://docs.langflow.org/get-started-quickstart ### Admin approval Documented: Role-based access control requires an authorisation plugin, which Langflow does not include. Without a plugin, Langflow allows every authenticated user. Evidence: https://docs.langflow.org/authentication-overview ### Data access & team sharing Documented with conditions: Reach depends on configuration. Webhooks require an API key by default; without authentication, calls are treated as sent by the flow owner. An MCP server without authentication is publicly reachable. The Shareable Playground, meant for testing only, makes a flow’s Playground usable at a public URL without installing Langflow or an API key. Evidence: https://docs.langflow.org/webhook Evidence: https://docs.langflow.org/mcp-server Evidence: https://docs.langflow.org/concepts-playground ## Prices and included usage Pricing checked: 2026-11-05 Langflow is available under the MIT licence for self-hosting. Model costs depend on your provider. According to DataStax, hosted Langflow was removed on 9 April 2026. ### Langflow (open source) No software fee stated Open-source software for self-hosting under the MIT licence. Price source: https://github.com/langflow-ai/langflow A pricing page at www.langflow.org/pricing could not be retrieved (404). Pricing details: https://www.langflow.org ## Operation & oversight Based on public provider documentation. Assessments are editorial inferences from the stated properties. Actual time requirements and reliability have not been measured. Deployment covered: Self-operated installation (desktop, package or Docker) Sources reviewed: 2026-11-05 ### Files and working state Documented: Flows, saved versions and chat messages are stored by default in the installation’s database (SQLite); LANGFLOW_DATABASE_URL selects another database. Evidence: https://docs.langflow.org/concepts-flows Evidence: https://docs.langflow.org/memory ### Sign-ins Documented with conditions: Official release packages and tagged source archives disable automatic login by default; development branches do not. For new Python package installations, the quickstart says to set LANGFLOW_AUTO_LOGIN=false with a strong superuser password; Docker images require authentication by default. On session lifetimes the reviewed pages say only that tokens are short-lived. Evidence: https://github.com/langflow-ai/langflow Evidence: https://docs.langflow.org/get-started-quickstart Evidence: https://docs.langflow.org/authentication-overview ### Starting further runs Documented with conditions: You start flows in the Playground, via the API or by webhook POST. The webhook response only confirms the start. The reviewed pages describe no general scheduler. Nodes are built and executed sequentially. Evidence: https://docs.langflow.org/webhook Evidence: https://docs.langflow.org/concepts-flows Evidence: https://docs.langflow.org/environment-variables ### Human intervention Open in reviewed sources: The reviewed pages describe no built-in human approval before individual tool calls. They name the Playground as the place for testing. Evidence: https://docs.langflow.org/agents Evidence: https://docs.langflow.org/concepts-playground ### Detecting errors Documented with conditions: Logs go to stdout by default, with ERROR as the default level; a file can be configured. Log retrieval through an endpoint is off by default. The MCP server’s default tool timeout is 180 seconds. A Parser component can fail to build if it receives no or bad data. The pages describe no alerting on errors. Evidence: https://docs.langflow.org/logging Evidence: https://docs.langflow.org/mcp-server Evidence: https://docs.langflow.org/webhook ### Resuming after failures Documented with conditions: Saved versions of a flow can be restored, optionally backing up the current draft. If a flow was saved elsewhere, Langflow refuses the save and offers update, duplicate or load latest. The reviewed pages do not describe resuming an interrupted run. Evidence: https://docs.langflow.org/concepts-flows ### Editorial assessment You operate Langflow yourself and are responsible for sign-in, database and logs. Set authentication for webhooks and MCP servers deliberately and plan your own approvals and error monitoring. ## Memory and conversation continuity Scope: Chat memory of the Agent component and the Message History component Sources accessed: 2026-11-05 Chat histories are stored in the Langflow database by default, grouped by session ID, and can be viewed, edited and deleted in the Playground. ### Storage and scope Documented: By default Langflow stores data in a SQLite database, with messages in the messages table. LANGFLOW_DATABASE_URL lets you use another database, including for chat memory only. - [Langflow: Memory](https://docs.langflow.org/memory) ### Selection and timing Partly documented: The Agent component has chat memory enabled by default and retrieves a limited number of earlier messages; the number is configurable, and the page gives no default. The pages do not describe how content is selected for lasting memory. - [Langflow: Agents](https://docs.langflow.org/agents) - [Langflow: Memory](https://docs.langflow.org/memory) ### Inspect, edit and delete Documented: In the Playground you open the message logs from a session’s options, edit or delete individual messages and delete whole sessions. Editing changes the messages table. - [Langflow: Playground](https://docs.langflow.org/concepts-playground) ### Agent and channel boundaries Partly documented: Histories are grouped by session ID; the default ID is the flow ID. For multiple users the documentation suggests custom session IDs. The page does not show whether messages are separated across projects. - [Langflow: Memory](https://docs.langflow.org/memory) ### Retrieval Documented: The agent can retrieve earlier messages of the same session. The Message History component reads data even outside a chat. For semantic search across sessions the page points to a memory base rather than chat memory. - [Langflow: Memory](https://docs.langflow.org/memory) ### Session reset and compaction Partly documented: A different session ID, in the editor or via the API, starts a separate session. According to the documentation the agent works with a rolling context window per session. The reviewed pages describe no new-chat button or compaction. - [Langflow: Playground](https://docs.langflow.org/concepts-playground) - [Langflow: Agents](https://docs.langflow.org/agents) ## Context A model provider key and a single flow to test with. Role-based access control needs a plugin that Langflow does not include; an MCP server without authentication is publicly reachable. Provider website, documentation, blog and repository reviewed. A pricing page could not be retrieved. Feature descriptions come from the provider; example tasks are editorial. Hands-on tests are still pending. ## Open points Europe, availability still unclear: - Europe: no provider statement ## Sources - https://github.com/langflow-ai/langflow - https://www.langflow.org - https://docs.langflow.org - https://docs.langflow.org/get-started-installation - https://docs.langflow.org/get-started-quickstart - https://docs.langflow.org/agents - https://docs.langflow.org/mcp-server - https://docs.langflow.org/mcp-client - https://docs.langflow.org/webhook - https://docs.langflow.org/memory - https://docs.langflow.org/concepts-flows - https://docs.langflow.org/concepts-playground - https://docs.langflow.org/authentication-overview - https://docs.langflow.org/deployment-overview - https://docs.langflow.org/logging - https://docs.langflow.org/environment-variables - https://docs.langflow.org/contributing-telemetry - https://www.langflow.org/blog/big-news-for-langflow - https://docs.datastax.com/en/langflow/index.html ## Provider The project is developed in the langflow-ai/langflow repository under the MIT licence. The blog names DataStax as parent company (as of February 2025). - Operator: Langflow - Organization: Open-source project - Website: https://www.langflow.org - Reviewed: 2026-11-05 - Repository: https://github.com/langflow-ai/langflow - GitHub-Stars: 156355 (2026-11-05) - License: MIT - Provider source: https://www.langflow.org - Provider source: https://github.com/langflow-ai/langflow - Provider source: https://www.langflow.org/blog/big-news-for-langflow [Sources and method](https://ki-agentenkompass.de/en/sources/)