Langflow is an open-source visual builder for agents and RAG flows. Its OpenAI component has an OpenAI API Base field that, in the component’s own words, lets you “use other APIs”. BroadNet Inference serves chat completions at https://inference.broadnet.ai/v1, so that component works with a typed BroadNet model ID. BroadRouter is BroadNet’s own site; this follows the Langflow 1.13.x docs and the component’s source in v1.13.2.
Setup
-
Add the component. In a flow, open the components panel and add OpenAI from the OpenAI bundle. Langflow 1.13.2 installs this bundle with the main package.
-
Model Name. Click the field and type
broadnet/glm-5.3. It accepts custom values, so you aren’t limited to the OpenAI models in the list. -
OpenAI API Key. Paste your BroadNet key, or pick a global variable that holds it.
-
Advanced settings. Open the component’s controls and set:
Field Value OpenAI API Base https://inference.broadnet.ai/v1Max Tokens 4000 -
Run it. Connect a Chat Input and Chat Output, open the Playground and send a message. A reply means the settings are right.
Two Langflow checks that can block BroadNet
Langflow 1.13 has SSRF protection on by default. Two rules matter:
-
Keys from the server environment. If the API key comes from the server’s environment (for example a global variable loaded from an environment variable), Langflow refuses to send it to a non-default endpoint unless the host is allowlisted. The operator sets:
export LANGFLOW_PROVIDER_CREDENTIAL_ALLOWED_HOSTS=inference.broadnet.aiA key typed into the component isn’t affected.
-
Private addresses. Loopback and private IPs are blocked unless they are in
LANGFLOW_SSRF_ALLOWED_HOSTS. BroadNet is a public HTTPS host, so this only matters if you route it through a local proxy.
Why not the OpenAI Compatible provider?
Langflow 1.13 also has an OpenAI Compatible model provider under Settings → Model Providers. Its docs say it validates the connection against the endpoint’s /v1/models route and offers the models it finds. When we checked on 16 November 2026, BroadNet’s /v1/models listed only broadnet/cohere-transcribe-arabic and broadnet/cohere-transcribe, so the chat models can’t be picked there. The OpenAI component’s typed model name avoids that.
Which model to pick
| Flow | Model |
|---|---|
| Agent component with tools | broadnet/glm-5.3 |
| Flows with images or Arabic text, with tools | broadnet/qwen3.8-27b |
| Long-document flows, flat-price plan | broadnet/deepseek-v4.1-flash |
| Trying Langflow on the starting credit | broadnet/gemma-4-26b-a4b (32,768-token context) |
BroadNet’s docs list tools and structured output for GLM 5.3 and Qwen3.8 27B only. The component’s JSON Mode sends response_format: {"type": "json_object"}; use it with those two models. Prices for every model.
How we tested this
We did not run the Langflow UI. On 2026-11-15 we installed lfx 1.13.2 and the lfx-openai bundle, whose OpenAI component file matches the v1.13.2 source, and built the component in Python against a local mock server that speaks BroadNet’s chat-completions shape. With a typed key and LANGFLOW_SSRF_ALLOWED_HOSTS set for the local mock, the request went to /v1/chat/completions with the model ID broadnet/glm-5.3 unchanged, a Bearer key, temperature: 0.1 and a seed. The Max Tokens value was sent as max_completion_tokens; BroadNet’s API reference describes max_tokens. A key copied from the server environment was refused until we also set LANGFLOW_PROVIDER_CREDENTIAL_ALLOWED_HOSTS. We had no BroadNet key, so we did not test real model output.
Troubleshooting
- “Refusing to send a server-provisioned API credential”. Add
inference.broadnet.aitoLANGFLOW_PROVIDER_CREDENTIAL_ALLOWED_HOSTS, or type the key into the component. - 401. The key is wrong, expired or out of allowance.
- 404. OpenAI API Base must end in
/v1, and Model Name must match BroadNet’s ID exactly. Paid models need a top-up; the $5 credit covers onlybroadnet/gemma-4-26b-a4band the two Transcribe models. - Empty replies. Reasoning counts toward the output limit on DeepSeek V4.1 Flash, GLM 5.3 and Qwen3.8 27B. Raise Max Tokens. Why answers come back empty.
- 429. A pay-as-you-go key allows 30 requests a minute and 4 in flight. See errors and retries.
