Providers

Rune Agent works with a set of model providers. A provider is a service that hosts models and authenticates your requests. Rune ships with built-in support for the major cloud providers, plus local models and custom OpenAI-compatible endpoints. Every model, whatever its provider, shows up in one flat list and is selected the same way.

This guide covers the hosted cloud providers: how to onboard them, what each one offers, and how to configure them in detail.

The built-in providers

Rune has six built-in cloud providers:

Provider	Sign-in	Models
OpenAI	API key	GPT family
Anthropic	API key	Claude family
Gemini	API key	Gemini family
Bedrock	Bedrock API key or your AWS credentials	Claude, Nova, Llama, Mistral, and DeepSeek families hosted on Amazon Bedrock
Codex	Sign in with ChatGPT (OAuth)	GPT-5 Codex family
Claude	Sign in with Claude (OAuth)	Claude family through your Claude Pro/Max subscription

The first four are the basic cloud providers: you onboard each by pasting an API key. Bedrock additionally accepts the AWS credentials you already use on the machine. Codex and Claude are browser sign-in providers. Codex signs in to ChatGPT and talks to the ChatGPT Codex backend on your behalf. Claude signs in to your Claude account and talks to Anthropic models through your Claude Pro/Max subscription.

note

Rune already knows the catalog of models each provider serves, so you do not declare them. Once a provider is authenticated, its models appear in the model list automatically.

Onboarding a provider

You onboard providers with the models command from the Rune console. The provider subcommands live under models providers:

models providers                # show all providers and their status
models providers <provider>     # show one provider's subcommands

How you authenticate depends on the provider.

API-key providers (OpenAI, Anthropic, Gemini, Bedrock)

OpenAI, Anthropic, Gemini, and Bedrock are onboarded by storing an API key. Each provider keeps a set of named keys, so you can hold more than one (for example a personal and a work key) and switch between them. The same four subcommands work for all four:

Command	What it does
models providers <provider> add <name>	Opens a masked prompt to paste a key, tests it against the provider, then stores it under <name>. The first key you add becomes the active one.
models providers <provider> use <name>	Make a stored key the active one. Takes effect immediately, no restart needed.
models providers <provider> status	List the stored key names and show which is active.
models providers <provider> remove <name>	Delete a stored key. If it was active, another stored key is promoted automatically.

To onboard OpenAI, for example:

models providers openai add work    # paste a key, store it as 'work'
models providers openai status      # confirm 'work' is the active key

The key is pasted into a masked prompt (each character shows as *, like a password field), verified with a live request to the provider, and only stored if the test succeeds. Keys are kept in Rune's local storage, not in your config file, so you never commit a secret. Anthropic, Gemini, and Bedrock work identically; just swap the provider name:

models providers anthropic add work
models providers gemini add work

Amazon Bedrock

Bedrock gives you Claude, Nova, Llama, Mistral, and DeepSeek models billed through your AWS account. It is the one provider with two ways to authenticate. Storing a Bedrock API key works exactly like the other API-key providers above, with one extra requirement: the key is bound to a single AWS region, so add takes the region alongside the key name:

models providers bedrock add work us-east-1

Without a stored key, Rune signs requests with the AWS credentials already on the machine, including SSO sessions, named profiles, and instance roles.

The Amazon Bedrock guide covers both paths in detail: signing in with IAM Identity Center (SSO), choosing a profile and region, private endpoints, and how model access shapes the model list.

Codex (Sign in with ChatGPT)

Codex does not use an API key. You sign in to your ChatGPT account and Rune uses that session to reach the ChatGPT Codex backend:

models providers codex login    # opens a browser to sign in to ChatGPT
models providers codex status   # show the signed-in account and its plan

login opens your browser to complete the OAuth flow, then stores the resulting credentials locally and refreshes them as needed. Once signed in, the Codex models appear in the model list like any other provider's.

Claude (Sign in with Claude)

Claude lets Rune Agent use Anthropic models through your Claude Pro or Claude Max subscription. You sign in with Claude; no Anthropic API key is required.

This is different from running Anthropic's standalone Claude Code CLI inside Rune. For the CLI integration, see the Claude Code guide.

Rune has two built-in ways to use Claude-family models:

Provider	Authentication	Billing / quota
anthropic/...	Anthropic API key	Anthropic API usage
claude/...	Claude OAuth sign-in	Claude Pro/Max Agent-SDK usage credit

Use anthropic when you want API-key billing. Use claude when you want to use your Claude subscription.

To sign in, run:

models providers claude login

Rune opens a browser for the Claude OAuth flow, stores the resulting credential locally, and refreshes it when needed. Check the account and plan with:

models providers claude status

Once authenticated, Claude models appear alongside the rest of your models. Use the claude/ provider prefix when a model name is ambiguous:

models
model claude/claude-opus-4-6

You can also make a Claude subscription model the default for new chats:

config.yaml

models:
  default: "claude/claude-opus-4-6"

config.star

"models": {
    "default": "claude/claude-opus-4-6",
},

Most users do not need any models.claude config. Authentication is managed by models providers claude login, not by config files. If you front the Claude subscription endpoint with a proxy, set a base URL override:

config.yaml

models:
  claude:
    base_url: "https://claude-proxy.internal"

config.star

"models": {
    "claude": {"base_url": "https://claude-proxy.internal"},
},

Reasoning effort for Claude-family models is shared with the Anthropic provider, so configure it under models.anthropic.reasoning_effort.

note

Claude subscription usage in Rune draws from Claude's Agent-SDK credit, which is separate from your Claude.ai chat limits. If that credit is exhausted, enable usage credits ("extra usage") in your Claude account under Settings > Usage.

Selecting a model

After a provider is authenticated, run models (or agent models) in the Rune console to see its models in the list. Because the same model name can be served by more than one provider, Rune identifies models as provider/name, for example openai/gpt-5.4, anthropic/claude-opus-4-6, claude/claude-opus-4-6, or codex/gpt-5.4. A bare name resolves when only one provider exposes it; otherwise prefix it with the provider.

Set the default model for new chats in your config:

config.yaml

models:
  default: "anthropic/claude-opus-4-6"

config.star

"models": {
    "default": "anthropic/claude-opus-4-6",
},

You can also switch a focused chat's model with the chatmodel command-prompt command. See the agent intro for the model-selection commands.

Configuration

Authentication is managed entirely through the models providers commands above, so the config file holds no secrets. What you configure under models is provider behavior: which base URL to talk to, how much the model should reason, and how reasoning is summarized.

These settings are optional. With a default install and an authenticated provider, every provider works without any models block at all.

Reasoning effort

OpenAI, Anthropic, Gemini, and Bedrock each take a reasoning_effort that controls how much the model thinks before answering. Higher effort trades latency and token cost for more thorough reasoning. The accepted levels are:

none  minimal  low  medium  high  xhigh  max

Not every model supports every level. reasoning_effort is a standing default applied to whichever model you are using: models that support the level honor it, and models that do not fall back to their own default without warning you. Leave it unset to always use the model's default.

config.yaml

models:
  openai:
    reasoning_effort: "high"
  anthropic:
    reasoning_effort: "high"
  gemini:
    reasoning_effort: "medium"
  bedrock:
    reasoning_effort: "medium"

config.star

"models": {
    "openai": {"reasoning_effort": "high"},
    "anthropic": {"reasoning_effort": "high"},
    "gemini": {"reasoning_effort": "medium"},
    "bedrock": {"reasoning_effort": "medium"},
},

You can also change effort per conversation with /effort in a chat, or set the session default with agent effort <level>. Unlike the config value, these are explicit per-session choices, so Rune warns you when the current model cannot honor the level you asked for.

Reasoning summary

models.reasoning_summary controls how Rune surfaces a reasoning model's thinking. It applies to OpenAI and Codex and is one of:

Value	Meaning
auto	Let the provider choose (the default).
concise	Show a short summary of the model's reasoning.
detailed	Show a fuller summary.
disabled	Do not show reasoning summaries.

config.yaml

models:
  reasoning_summary: "concise"

config.star

"models": {
    "reasoning_summary": "concise",
},

Base URL overrides

Each provider talks to its standard endpoint out of the box. You can override the base_url to route through a proxy, gateway, or compatible alternative endpoint, while still using that provider's authentication and model catalog.

config.yaml

models:
  openai:
    base_url: "https://openai-proxy.internal/v1"
  anthropic:
    base_url: "https://anthropic-proxy.internal"
  gemini:
    base_url: "https://gemini-proxy.internal"
  codex:
    base_url: "https://codex-proxy.internal"
  claude:
    base_url: "https://claude-proxy.internal"

config.star

"models": {
    "openai": {"base_url": "https://openai-proxy.internal/v1"},
    "anthropic": {"base_url": "https://anthropic-proxy.internal"},
    "gemini": {"base_url": "https://gemini-proxy.internal"},
    "codex": {"base_url": "https://codex-proxy.internal"},
    "claude": {"base_url": "https://claude-proxy.internal"},
},

Leave base_url unset to use the provider's default. For Codex, the default is the ChatGPT Codex backend; for Claude, the default is Anthropic's Claude subscription endpoint using Claude Code-compatible request headers. Override either only if you front it with a proxy. Bedrock also takes a base_url for VPC endpoints and gateways; see the Amazon Bedrock guide.

note

To point at a fully self-hosted or third-party OpenAI-compatible server (where you declare your own models and key), use a custom provider instead. Base-URL overrides here keep the built-in provider's authentication and known model catalog.

Configuration reference

Key	Type	Description
models.default	string	Default model for new chats, as provider/name.
models.reasoning_summary	string	Reasoning summary level for OpenAI and Codex: auto, concise, detailed, or disabled.
models.openai.base_url	string	Override the OpenAI API base URL.
models.openai.reasoning_effort	string	Reasoning effort for OpenAI models.
models.anthropic.base_url	string	Override the Anthropic API base URL.
models.anthropic.reasoning_effort	string	Reasoning effort for Anthropic models.
models.gemini.base_url	string	Override the Gemini API base URL.
models.gemini.reasoning_effort	string	Reasoning effort for Gemini models.
models.bedrock.profile	string	Named AWS profile to sign requests with when no Bedrock API key is stored.
models.bedrock.base_url	string	Override the Bedrock runtime endpoint, for VPC endpoints and gateways.
models.bedrock.reasoning_effort	string	Default reasoning effort for Bedrock models that support extended thinking; individual chats override it with chateffort or /effort.
models.bedrock.cache_control	string	Set to default to cache the stable part of each request and cut the cost of repeated context.
models.codex.base_url	string	Override the Codex backend base URL.
models.claude.base_url	string	Override the Claude subscription backend base URL.

See also

• Amazon Bedrock: both Bedrock authentication paths in detail, including SSO sign-in and private endpoints.
• Custom Model Provider: connect any OpenAI-compatible endpoint with your own model catalog.
• Local Models: run GGUF models on your own machine with no API key.
• Claude Code: run Anthropic's standalone CLI inside Rune.
• Rune Console: the models command and its subcommands.