mohManual

Manual/Get started

Choosing an AI model

Chapter 04 · 3 min read

A model is the AI that does the thinking; a provider is the company or program that runs it. moh works with many of them, and you can connect several at once.

Connect a provider

moh provider add

A short guided setup starts. Its first question decides the rest:

I have an API key

A key is a secret code from the provider's website. Paste it once; moh tests it with one small request and stores it in your personal settings — never inside a project.

I have a subscription

Claude Pro/Max, ChatGPT Plus/Pro or a personal Google account. moh opens your browser to log in — no key to copy, and your existing plan is what gets used.

If the login page won't open

On the login screen, press c to copy the full web address and paste it into your browser yourself.

Later, from the terminal:

$ moh provider status          # who is connected, and plan usage
$ moh provider login <name>    # log in again to a subscription
$ moh provider logout <name>   # forget a subscription login

Who you can connect

Built in: Anthropic, OpenAI, Google, GitHub Copilot, OpenRouter, Kimi, xAI and OpenCode, plus ready-made profiles for DeepSeek, Groq, Cerebras, NVIDIA NIM, Together AI, Fireworks AI, Hugging Face, Mistral, Moonshot, MiniMax, Z.ai, Qwen, Xiaomi MiMo, Vercel AI Gateway, Cloudflare AI Gateway and Baseten. Programs that run AI on your own computer — like Ollama or LM Studio — connect as a "compatible endpoint".

Switch model

Type /model (or press ctrl+l) to open the model picker. The new choice applies from your next message. In the picker, r refreshes the list from the provider, so models released yesterday appear without waiting for a moh update.

Good to know

If your conversation is already longer than the new model can hold, moh refuses the switch rather than breaking the session — and offers to summarise the conversation first (see long conversations).

Thinking levels

Some models can "think" longer before answering. Press ctrl+y to cycle the level — off, low, medium, high, and on some models xhigh or max — or open /thinking. Levels a model doesn't support are shown as unavailable, never silently changed. Higher levels are slower and usually cost more.

The safety net: automatic fallback

Connect more than one provider and moh builds a fallback chain for you. If the AI you picked is out of quota, overloaded, offline, or answers with nothing, the same request is retried on the next one — and the conversation says so.

fallback
✕ anthropic/claude-sonnet-4 · 429 rate limited
↪ fallback → openrouter/deepseek-chat
model anthropic/claude-sonnet-4 → openrouter/deepseek-chat
Your choice stays selected; the status bar shows "selected → serving" while another model stands in.
  • Each connected provider takes part with its own preferred model.
  • In the settings (ctrl+s → Fallback models) you can pick that model or leave a provider out of the chain with x.
  • moh never invents a substitute: every stop is a model you configured.

How much you are using

Press ctrl+q for usage: what your provider reports about your plan (where it reports anything), and a local tally of messages and tokens from your own conversations. /session shows a report of the current conversation, with an estimated cost where prices are known. For longer periods, see usage reports.

Tip

Tokens are the pieces of text an AI reads and writes — roughly ¾ of a word each. Providers bill by tokens; models with a bigger "context window" can hold more of them at once.

Written for moh 0.61. Inside moh, ctrl+h opens the built-in manual for your exact version.