> ## Documentation Index
> Fetch the complete documentation index at: https://docs.assistable.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Configure flow model access

> Choose a provider and model and understand which connected credentials the flow uses.

Conversation and Extraction nodes need working model access. Configure the provider and model in **LLM Config**, then test the actual flow before publishing.

## Choose a model

1. Open the flow in the intended subaccount and select **LLM Config**.
2. Review the connected keys under **Subaccount API Keys**.
3. Select the provider and a model offered by the current selector.
4. Check any node-specific model overrides as well as the flow default.
5. Wait for the draft to save and run a representative conversation and extraction test.

The provider selector includes OpenAI, Anthropic, and OpenRouter. Available models and working credentials must agree; choosing a provider name alone does not grant access to its models.

## Use your own credentials

The panel provides connection controls for OpenAI and Anthropic integrations. Connect the provider you intend to use and check the connection result. These are shared subaccount integrations, so changing or disconnecting a key can affect other flows in that subaccount.

Do not paste a provider secret into a prompt, variable intended for customer output, or website embed. A provider key is different from an Assistable API key used with the v3 API. Keep both private.

## Platform access and overrides

The runtime checks configured flow credentials before falling back to available platform model access. Node overrides can change the requested model. Platform fallback can resolve to a different supported provider or model, so a selector value alone is not proof of which model produced a reply.

Test the answer quality and timing you need rather than assuming every selected model is available. If exact model identity is a requirement, confirm it using the test result details and current configuration.

## Troubleshooting

| Symptom | What to check |
| - | - |
| No API key configured | Connected integration and availability of platform access |
| Authentication error | Correct provider, valid key, and revoked or expired credentials |
| Model unavailable | Model name and the provider account's access |
| Rate limit | Provider quota and request frequency |
| A node behaves differently | Its model override and prompt, not only flow defaults |

After changing credentials or models, retest the affected paths and publish the intended flow version. See [testing](/v3/guides/flows/test).

## Model configuration reference

| Control | Options and constraints |
| - | - |
| Subaccount API Keys | Connect or disconnect OpenAI and Anthropic. OpenRouter is listed as **Coming soon** in the connection section; selecting it elsewhere does not establish credentials. |
| Default Provider | OpenAI, Anthropic, or OpenRouter. New configuration starts with OpenAI; actual access depends on credentials/platform availability. |
| Default Model | Choose an entry returned for the selected provider. A blank selection is not a promise of a particular model. |
| Temperature | `0`–`2`, in `0.1` increments; default `0`. Higher values request more variation, subject to the model’s capabilities. |
| Fallback Provider | None, OpenAI, Anthropic, or OpenRouter. None is the editor default. |
| Fallback Model | Available after choosing a fallback provider. It must be usable with that provider’s credentials. |

Conversation nodes can override **Provider**, **Model**, **Temperature**, **Fallback Provider**, and **Fallback Model**. Unset values inherit the flow configuration. The node also exposes **Priority Processing** for OpenAI/OpenRouter; it inherits the configured value or defaults off. This requests the provider’s priority mode and can change usage cost. It does not guarantee a response time. Anthropic does not support this control here.

## Which models appear

| Access path | Selector behavior |
| - | - |
| Platform access without a connected OpenAI key | Offers **Kimi K2.6** and/or **GPT 6 Luna** when the corresponding platform service is available. |
| Connected OpenAI key | Loads the provider’s accessible model list and filters to supported GPT/chat and o1/o3 families. The list can vary by provider account. |
| Anthropic | Lists **Claude Haiku 5.5** and **Claude Haiku 4.5**. Listing a model does not verify your credentials. |
| OpenRouter | Retrieves its model catalogue when credentials are available; the editor does not currently provide self-service connection here. |

The provider-backed lists are loaded dynamically, so the selector is the complete list available to your configuration at that time. Model overrides and fallback can change the model used for a particular turn; inspect the model/provider shown in expanded test details.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.