TypeClawAdd a provider
Give the agent more than one model — a cheaper one for routine work, a stronger one for hard problems
You picked one model provider during typeclaw init. That's enough to run — but you'll often want more than one: a fast, cheap model for routine chatter and a stronger, pricier one for the hard problems. TypeClaw lets you wire up several providers and point different kinds of work at different models.
Add another provider
Adding a provider is one command. Pick the one you have an account with:
typeclaw provider add anthropic --key sk-ant-...
typeclaw provider add openai-codex --oauth # log in with a ChatGPT subscription
typeclaw provider add fireworks --key fw_...
typeclaw provider add zai --key ... # Z.AI pay-as-you-go
typeclaw provider add zai-coding --key ... # Z.AI GLM Coding Plan
typeclaw provider add xai --key xai-... # xAI (Grok) — also supports --oauth
typeclaw provider add minimax --key ... # pay-as-you-go API key or Token Plan key (sk-cp-…)
typeclaw provider add deepseek --key sk-... # DeepSeek pay-as-you-goThe supported providers are openai, openai-codex, anthropic, fireworks, zai, zai-coding, xai, minimax, and deepseek. The key (or browser login) is stored in secrets.json, git-ignored like the rest of your credentials.
MiniMax: pay-as-you-go or Token Plan
MiniMax sells one minimax provider under two billing surfaces — pay-as-you-go API keys and Token Plan Subscription
Keys (sk-cp-…). Both are saved to the same MINIMAX_API_KEY slot and work against the same endpoint; typeclaw init asks which one you have only so it can link you to the right dashboard.
xAI (Grok): API key or subscription
xai is dual-auth like anthropic: pass --key xai-… for the developer API (XAI_API_KEY), or --oauth to log in
with a Grok subscription. When both exist, the on-disk OAuth credential wins — run typeclaw provider remove xai to
fall back to the key.
Use a subscription you already pay for
openai-codex, anthropic, and xai support --oauth — log in through the browser with an existing ChatGPT,
Claude, or Grok subscription instead of pasting an API key.
Point work at the right model
Models are addressed through profiles — named slots the agent reaches for depending on the job. The main one is default; there are also conventional fast, deep, and vision slots.
typeclaw model set default openai/gpt-5.4-nano
typeclaw model add deep anthropic/claude-opus-4-8
typeclaw model add fast fireworks/accounts/fireworks/routers/kimi-k2p6-turboNow routine turns run on the cheap default, and work that asks for deep (like a code review) gets the heavyweight model. Run typeclaw model list --available to see every model ref you can use.
Going deeper
Next: Add a channel — open the agent up to your team in Slack, Discord, or beyond.