Configuring LLM Providers

Cremind talks to language models through providers. A provider is a named integration — anthropic, openai, groq, copilot — that holds the credentials and the list of models Cremind discovered for it. You configure providers once, then assign their models to the high / low / default model groups the agent reaches for.

Everything on this page has a control on the Settings -> LLM Providers page in the web UI, and a one-to-one counterpart in the cremind llm CLI. Anything you change in one shows up in the other.

In the web UI

Sidebar -> Settings -> LLM Providers

The page lists each provider as a card showing whether it is configured, its active auth method, and its model count. The card's Configure panel sets the key and auth method, and the Model groups section at the top of the page assigns models to groups.

Configuring a key-based provider

Anthropic, OpenAI, and Groq authenticate with an API key. In the UI, open the provider's card, paste the key, and save. From the CLI:

cremind llm providers configure anthropic --api-key sk-ant-...

You can also switch the active auth method or pass extra provider fields (such as an organization id or a custom base URL) as JSON:

# Extra fields via JSON, merged into the body
cremind llm providers configure openai \
  --json '{"organization":"org-...","base_url":"https://api.openai.com/v1"}'

At least one of --api-key, --auth-method, or --json must be supplied. Validation is server-side: if the provider does not accept a field, the call returns an error and nothing is changed.

Listing providers and their status

cremind llm providers list

NAME       DISPLAY              CONFIGURED  MODELS  ACTIVE_AUTH
anthropic  Anthropic            yes         8       anthropic
openai     OpenAI               no          0
copilot    GitHub Copilot       yes         12      oauth_personal

Model discovery

Once a provider is configured, Cremind enumerates the models it exposes. List them for a single provider:

cremind llm providers models anthropic

ID                            NAME
claude-opus-4-7               Claude Opus 4.7
claude-sonnet-4-6             Claude Sonnet 4.6
claude-haiku-4-5-20251001     Claude Haiku 4.5

GitHub Copilot via device-code sign-in

GitHub Copilot does not use a pasted API key. Instead it uses a device-code OAuth flow: Cremind hands you a short code and a URL, you authorize in the browser, and the token is stored server-side. In the UI this is the Sign in with GitHub Copilot button; from the CLI it is two commands.

# 1. Start the flow — note the URL, the user code, and the device code
cremind llm device-code start

verification_uri  https://github.com/login/device
user_code         ABCD-1234
device_code       4fe...e8c
expires_in        900
interval          5

Open the verification_uri, enter the user_code, then poll with the device_code. The poll command blocks until you finish authorizing in the browser:

cremind llm device-code poll 4fe...e8c

complete (token stored server-side)

The device code is short-lived (expires_in seconds, typically around 15 minutes). If it expires before you authorize, run cremind llm device-code start again for a fresh code. Pressing Ctrl-C during the poll is safe and does not invalidate the code.

Assigning providers to model groups

Configuring a provider is only half the job — the agent does not use a model until it is assigned to a model group. Groups are the indirection that lets the agent reach for a heavyweight model when it plans and a cheaper model for routine tool calls.

# See the current high/low/default assignments
cremind llm model-groups get

{
  "default_provider": "anthropic",
  "model_groups": {
    "high": "anthropic/claude-opus-4-7",
    "low":  "anthropic/claude-haiku-4-5-20251001"
  }
}

# Point the high group at a stronger reasoning model
cremind llm model-groups set --high anthropic/claude-opus-4-7

Model groups have their own page that explains the high / low split and why it lowers token cost — read it next.

Removing a provider's configuration

To clear every stored field for a provider — API key, OAuth tokens, base URL, everything:

cremind llm providers delete-config openai

The provider stays known (it still appears in providers list), but its configured flag flips back to no and any models that depended on the config become unavailable.