Guide

OpenClaw + Langflow: Design Visually, Call It as a Tool

Q: How does an agent call a Langflow flow?

The agent posts to Langflow’s /api/v1/run/ endpoint with an input value and the flow’s input and output types, using an authenticated connector.

Q: Do I need the Langflow UI running in production?

The Langflow deployment must be running, but the agent depends only on the flow ID and its declared inputs. Editing a live flow can break the integration.

Q: Is authentication required?

Yes. Enable Langflow’s authenticated API and use a scoped key, because an unauthenticated deployment lets anyone execute flows that hold your model credentials.

Langflow is a visual builder that deploys flows behind an authenticated API. Connected to OpenClaw or Hermes, a flow you drew this morning becomes something a user can trigger from Telegram this afternoon.

What the connection actually is

Langflow's value is the canvas — you can see the retrieval step, the branch, the tool call. But the thing your agent talks to is much smaller: one HTTP endpoint, one flow ID, one declared input shape. Treat the flow as an API with a versioned contract and the integration stays boring. Treat it as a live document anyone can edit and it will break at the least convenient moment.

1. Deploy Langflow with its authenticated API

Follow the Langflow API documentation, enable authentication, and pin the version. Put it behind TLS. Then confirm the deployment answers and note the flow ID you intend to expose:

curl -fsS https://langflow.example.com/api/v1/version

2. Connect it from Dashboard Tools

Open Dashboard Tools, select a running instance, and connect the Langflow card. The placeholder endpoint is https://langflow.example.com; supply the scoped key your deployment requires. The connector installs a small HTTP client and a managed skill so the agent knows the call shape. The client lives at a managed path — tools/dashboard-api-connectors/langflow/tool — and the examples below use tool as shorthand for it.

# Verify the deployment
tool get /api/v1/version

# Run one flow
tool post /api/v1/run/FLOW_ID '{"input_value":"Hello","output_type":"chat","input_type":"chat"}'

The input_type and output_type fields are not decoration — they tell Langflow which component in the graph receives the input and which one's output to return. Getting them wrong produces an empty or unexpected result rather than an error, which is a confusing failure to debug from the agent side.

3. Pin the flow, then change it deliberately

Give the agent one flow ID per capability and write down what that flow promises to accept and return. When the flow needs to change in a way that alters its inputs or output shape, publish a new flow and switch the ID, rather than editing the one your users are currently hitting. Langflow makes editing easy; that is exactly the risk.

4. Keep the boundary honest

The client sets a 120-second socket timeout, which bounds a stalled connection rather than total runtime. Long-running flows need an async pattern and a deadline enforced by Langflow, not a longer wait.
Printed responses cap at 5 MB; have the flow return a summary and a reference rather than a whole corpus.
Model keys live in Langflow, not in the agent. Do not duplicate them into the instance config.
If the flow can write to external systems, require confirmation on the agent side before it calls.

Production hardening

Authenticate and scope the key to the instance; revoke in Langflow when an instance is retired.
Rate-limit at the proxy — a messaging channel can send far more requests per minute than a UI ever did.
Log flow IDs and outcomes, not inputs that may contain user personal data.
Keep a staging deployment. Testing flow edits against the endpoint your users hit is not testing.

Langflow versus Dify versus LangChain

Langflow suits teams who want the graph visible and deployable in one step. Dify is stronger when you want applications, publishing, and observability as first-class concepts. LangChain wins when the logic belongs in version control and code review. From the agent's perspective all three are the same thing: an authenticated HTTPS endpoint it calls as a tool.

Frequently asked questions

How does an agent call a Langflow flow?

By flow ID, over Langflow's authenticated run API. The connector posts to /api/v1/run/<flow-id> with an input value and the input and output types the flow expects, then returns the result to the agent.

Do I need the Langflow UI running in production?

The deployment needs to be running, but the canvas is a design surface, not the runtime contract. What your agent depends on is the flow ID and its declared inputs — which is why an unannounced edit to a live flow is the most common way this integration breaks.

Is authentication required?

Yes. Enable Langflow's authenticated API and connect with a scoped key. An unauthenticated Langflow instance lets anyone who finds it execute your flows — and your flows hold your model keys.