Connect LangBot

Use an existing LangBot installation with permission to manage models. These steps follow the current provider-based interface; older versions combine the provider, URL and key into a single model form.

1. Add a provider

Open the Models panel and select Add Provider. Fill in:

Field	Value
Provider Name	A unique local label, such as my-api
Provider Type	OpenAI
Request URL	The service Base URL above, including /v1
API Key	Your service key, entered privately without Bearer

The reviewed OpenAI requester has the internal ID openai-chat-completions. It uses OpenAI-compatible Chat Completions, including for compatible models from other vendors. Keep the URL at the API base; do not append /chat/completions.

Select Save. Keep existing providers and model assignments.

2. Add the models you need

On the saved provider’s card, choose Scan Models. Select the text models you want and choose Add Selected. Alternatively, use Add Model → Manual → Chat and enter an exact Model Name from the authenticated models endpoint.

Only enable Function Call for models with verified tool-calling support, and leave unverified vision capabilities off. Do not infer a context length from a model’s display name.

3. Assign the model to a pipeline

Open Pipelines and the pipeline you want to use. In AI Feature (called AI Capabilities in older documentation), inspect its Runner and select the added text model in that runner’s model setting. Save the pipeline. Keep its other prompts, tools and bot bindings unchanged.

The runner must use LangBot’s configured models. A runner backed by an external workflow platform takes its model settings from that platform instead; adding a LangBot provider does not change those settings. Current runner plugins can present different configuration fields, so follow the installed runner’s model selector. Do not rely on the first added model becoming the default for an existing pipeline.

Open the pipeline’s Debug Chat and send one short prompt. Model tests and chat prompts send paid requests; agent loops can make several calls. Confirm the selected model in your service dashboard before using the pipeline with a messaging bot.

Provider and pipeline settings persist after saving. Reopen the pipeline’s model selector if its list has not refreshed. Adding this connection does not send a message to an existing bot by itself.

For 404, check the /v1 suffix. For missing models, check the key and whether scanned models were added. Enabling a model does not grant tool permissions or configure a messaging channel.

See the reviewed provider form, OpenAI requester, and official model and pipeline documentation.