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

# Connecting an LLM provider

> Connect OpenAI or Anthropic and choose which models your agents and chat can use.

Erstan agents and Chat run on a large language model (LLM), so your workspace needs at least one connected AI provider before anyone can chat or run an agent. You connect a provider once with a workspace API key, then pick which of its models are available.

<Info>Connecting and managing an LLM provider is an Owner/Admin task. Members see the connected models in Chat but cannot change keys or the model allowlist — some controls show **Requires Admin permission to change.**</Info>

## Why a provider is required

Every reply in Chat and every AI step in an agent is generated by a model from a connected provider. Until one is connected, the chat composer is blocked and prompts you to connect a provider, with a button that links straight to **Connectors**.

Providers appear in the [Connectors](/connectors/overview) list with an **LLM** type badge. Erstan supports **OpenAI** and **Anthropic**.

## Connect a provider

You connect a provider by pasting one workspace API key from that provider. The key is stored for this workspace only.

<Steps>
  <Step title="Open the provider connector">
    Go to **Connectors**, then open **OpenAI** or **Anthropic**. (NetSuite is connected separately — see [Connecting NetSuite](/connectors/netsuite).)
  </Step>

  <Step title="Go to the Tokens tab">
    Open the **Tokens** tab and find the **Provider API key** field.
  </Step>

  <Step title="Paste your API key and save">
    Paste the API key from your OpenAI or Anthropic account and save. One workspace API key applies per provider.
  </Step>

  <Step title="Enable the models you want">
    Switch to the **Models** tab to choose which models agents and Chat can use (next section).
  </Step>
</Steps>

<Warning>
  The API key bills usage to your provider account, so treat it like a secret. Anyone the key reaches can spend against it. Use **Replace key** to rotate it and **Remove key** to revoke access for the whole workspace. Keys are scoped to this one workspace.
</Warning>

## Choose which models are enabled

The **Models** tab lists the provider's **Available models**, which are managed by Erstan Admin. From there you control your workspace allowlist:

* Toggle individual models on or off, then **Save**.
* Use **Reset to defaults** to return to the standard set.
* Search the model list and open a model for its details.
* Enabled models become the choices in the Chat model picker; Chat and agents fall back to the workspace default when no model is chosen.

<Tip>Keep the allowlist focused. Enabling only the models your team actually uses keeps the Chat model picker short and makes costs easier to predict.</Tip>

### Provider Tools

OpenAI and Anthropic also expose **Provider Tools** — provider built-ins such as web search, file search, and code interpreter — in a separate sub-card on the connector. These are managed alongside the provider rather than as NetSuite tools.

## How model choice surfaces in Chat

Once a provider is connected and models are enabled, you pick a model per conversation from **Model settings** in the chat composer's **+** menu. It sets the **provider, model, and reasoning effort** together.

<Frame caption="Choosing a model and reasoning effort">
  <img src="https://mintcdn.com/erstan/VvzUuagDdyVkLVu5/images/chat-model-settings.png?fit=max&auto=format&n=VvzUuagDdyVkLVu5&q=85&s=cb1391a1b106885a619a02055a38657a" alt="The chat model picker with provider (Anthropic, OpenAI), model list, and reasoning effort" width="2880" height="1800" data-path="images/chat-model-settings.png" />
</Frame>

* Chat remembers your **last-used model** so you don't reset it every time.
* If you don't choose one, Chat uses the workspace default.
* Higher **reasoning effort** trades speed for more careful answers on complex finance questions.

For the full Chat controls — connections, knowledge, and web search alongside the model — see [Knowledge, tools, web search & models](/chat/knowledge-tools-models).

## Common questions

<AccordionGroup>
  <Accordion title="Do I need a provider if I already connected NetSuite?">
    Yes. NetSuite is the data connector; an LLM provider supplies the model that reads and reasons over that data. Chat and agents stay blocked until both are in place. See [Connect NetSuite](/connectors/netsuite).
  </Accordion>

  <Accordion title="Can each member use their own API key?">
    No. There is one workspace API key per provider, set by an Owner or Admin, and everyone in the workspace shares it.
  </Accordion>

  <Accordion title="A model I want isn't in the list.">
    Available models are managed by Erstan Admin. If a model isn't offered for the provider, it can't be enabled. Check that your provider account has access to that model, and confirm an admin hasn't turned it off in the **Models** tab.
  </Accordion>

  <Accordion title="Chat still says it needs an LLM provider.">
    The provider has no valid key, or no models are enabled. Open the connector, confirm the **Provider API key** is saved on the **Tokens** tab, and enable at least one model on the **Models** tab.
  </Accordion>
</AccordionGroup>

## Next steps

<CardGroup cols={2}>
  <Card title="Connectors overview" icon="plug" href="/connectors/overview">
    See every connector and its connection status.
  </Card>

  <Card title="Connect NetSuite" icon="database" href="/connectors/netsuite">
    Link your NetSuite account so agents can read and act on your data.
  </Card>

  <Card title="Knowledge, tools & models in Chat" icon="sliders-horizontal" href="/chat/knowledge-tools-models">
    Control what Erstan can see and which model it uses per message.
  </Card>

  <Card title="Connection troubleshooting" icon="wrench" href="/connectors/troubleshooting">
    Fix expired keys, unhealthy connections, and re-authentication.
  </Card>
</CardGroup>