Think · conversation · updated 22 October 2026

Chat

Ask ALMA anything, with a picture or a voice note if the model can take one. 13 models are free; the rest cost a few credits a message, shown before you send.

See the recipes Models for /chat
  • 142chat models
  • 13free models
  • 1–25credits a message on paid models
  • 12recent messages remembered per channel

Entry I

Start here

Run /chat with your message (up to 2000 characters). Leave model empty to start on ALMA, the free model, or type to search. A panel opens where you can still change the model (pick a family, or press Find… to search by name, provider or id) before pressing the Ask button, which shows the price.

Optional: image for models marked img (93 of them), audio for models marked audio (Gemini 4 Argon, Meta Muse Wonder and Voxtral Small), and hidden:true to keep it to yourself.

Entry II

Web search and memory

Web: leave web empty and ALMA decides whether the question needs a live search; true always searches, false never does. The panel's Web button cycles Auto, On and Off. Up to 2 links in your message are read too, unless web is off. Answers that used the web have a Sources button.

Memory: ALMA reads the last 12 messages of your conversation in that channel. Leave continue empty and it decides whether your message follows on; false starts fresh. Chat memory is kept for up to 30 days, and /privacy delete removes it.

These automatic choices never override a setting you picked, and never change the price.

Entry III

Think and Detail

35 models (Claude, GPT, Grok and a few others) add a Think menu: how much the model reasons before answering. Default keeps the model's own setting. Higher effort helps hard problems and costs more: the message price is multiplied by 1 at Low, 2 at Medium, 2 at High, 3 at Extra high and 3 at Max. Thinking is billed as output, so easy questions do not need it.

Some GPT models also have Detail: Short, Medium or Long replies.

Entry IV

Asking well

Give the context

The model knows nothing about you or your project. Say who it is for, what you already tried and what a good answer looks like.

Say the format

"Three bullet points", "a table with price and weight", "under 100 words". The model will follow it.

One question at a time

A follow-up with Continue keeps the thread, so you can go step by step.

Paste, don't describe

Paste the error, the paragraph or the numbers. Attach a screenshot to a model that reads pictures.

Match the model to the job

A free model for quick questions and drafts; a paid model with Think for code, maths and careful reasoning.

Entry V

After the answer

Long answers are split into pages with Prev and Next. Continue opens a small panel to ask a follow-up (Ask…), switch model or web, or hand the conversation to Image, Video, Speech or Music to make something from it.

Entry VI

Every option

What you can type after /chat, read from the command itself. You can run it again after 5 seconds.

OptionWhat it doesTakes
message needed Your message textup to 2,000 characters
model Starting model — you can switch on the next panel (type to search) texttype to search
image Optional image for models that support image refs (img) file
audio Optional audio to transcribe (models marked audio) file
web Live web search (omit = ALMA decides; true = always; false = never) true / false
continue Keep this channel’s chat history (omit = ALMA decides; false = start fresh) true / false
hidden Only you see the confirm / reply true / false

Entry VII

Recipes

Requests you can copy into Discord and change to fit.

A quick fact check

Copy this

/chat message: What changed in the latest Discord.js major release? Give me the five changes most likely to break a bot. web: true

Why it works. A current question with web forced on, and a clear format.

Debug with context

Copy this

/chat message: My Node bot crashes with "Cannot read properties of undefined (reading 'id')" when someone uses a button after a restart. Here is the handler: (paste). What is wrong and how do I fix it?

Why it works. The error, the code and when it happens. Pick a paid model and raise Think for code.

Read a picture

Copy this

/chat message: What plant is this, and is it safe for cats? image: (a photo) model: (one marked img)

Why it works. A picture needs a model marked img.

Entry VIII

Tips

  1. I

    Start on the free model; switch in the panel only when the answer needs more.

  2. II

    Raise Think for maths, code and planning. Leave it on Default for chat.

  3. III

    Use continue:false to change subject; old messages in the channel would otherwise colour the answer.

  4. IV

    Paste links: ALMA reads the pages when web is not off.

Entry IX

Limits

  • Messages are up to 2000 characters.

  • Up to 2 chats at once, 5 seconds between /chat commands, and 4 minutes per answer.

  • Free models are shared and can be slow or busy; a paid model is the fix when it matters.

  • A picture or audio file is only read by models marked for it.

Entry X

Questions

Which model is "ALMA"?

The free default. It routes your message to a free model on OpenRouter, with a list of free chat models as a fallback.

Why did it not search the web?

With web on Auto, ALMA skipped a search it judged unnecessary. If the answer looks out of date it is answered again with a search; or set web:true.

Does Think change the price?

Yes, on models that have it: ×2 at Medium, ×2 at High, ×3 at Extra high, ×3 at Max. Default and Off keep the base price.

Can it remember me across channels?

No. Memory is per person and per channel.

/decisionDecision /digestDigest /wikiWiki

Where the advice comes from

Prompting advice is from these free public guides, matched to what ALMA lets you set. Prices and limits are read from the bot itself, as of 22 October 2026.

Try it

Type /chat in any server with ALMA, or in a direct message to the bot.

All guides

Something here wrong, or missing? Tell us.