Model configuration

This page covers how to choose which AI model does your work in Multilo: letting Auto decide, picking a model yourself, setting models for agents, and the settings that control when Multilo switches models for you. For what models are and why they differ, see how AI models work.

How a model is chosen

Whatever you choose, Multilo only ever uses models your plan includes.

Set the model

Auto

Open the model menu at the bottom of the chat and choose Auto at the top of the list. Multilo then picks a model for each message — fast for simple asks, stronger for demanding or accuracy-critical ones, leaning toward capable models whenever your sources or citations are involved. New accounts start on Auto. See Auto model selection.

Pick a model yourself

Choose any model from the same menu. That turns Auto off for the chat until you switch it back. Pick a model yourself when you want:

• consistency — the same model every time, for work you may need to reproduce;
• a comparison — two models on the same task;
• a preference — a model you simply like for this work.

A setting you change mid-chat applies from your next message.

Models for agents

Each agent has a sensible default model, and you can override it per run: a fast model to iterate, then a stronger one for the final pass. See models.

Which models you can use

The list of models changes as new ones arrive, and the model menu always shows what’s available to you. The pricing page names the current models per plan. See feature availability.

Read the model picker

Each model in the menu carries a short badge:

For which tier suits which task, with sample prompts, see choosing a model.

Reasoning depth

Two independent levers decide how deeply the AI works on a task:

• The model tier — a more capable tier reasons better;
• An agent’s MODE — a light pass or a deep one, whatever the model.

They multiply: a flagship model with a deep MODE for the hardest work, a fast model with a light MODE for a quick pass. Deeper reasoning costs more credits and takes longer, so save it for work that needs it.

When Multilo switches models

• The model can’t do the job. If the model you picked can’t handle a request — it needs a tool the model doesn’t support, or the request is bigger than its context window — Multilo sends that one turn to a capable model on your plan and tells you. Your choice stays in place for the next turn.
• A message is flagged. If a model flags a message, the setting Switch models when a message is flagged decides what happens: Switch automatically to a model in the same family, or Ask each time.
• Credits run low (Auto). Auto shifts to a lighter model so you can keep working instead of hitting a wall.
• A stronger model would help (Auto). Auto uses the best model your plan includes and tells you when a stronger one is available.

See which model answered

With Auto, every answer carries an “Answered by…” line; hover it to see why that model was chosen. Next to it, Retry with a stronger model re-runs the message on a more capable model when one is available to you. For a summary of every model Auto used in the current session, with the reason for each — handy for reproducibility — type:

/models

Context window and compaction

Each model can read a limited amount of text at once — its context window. Models marked Long ctx hold more. In long runs, Multilo compacts older material automatically so the agent keeps its thread. To cap how much one task can spend, set a per-task token budget in Settings → AI; it’s off unless you set one, and when a run reaches it, the agent pauses and asks before going on.

Settings at a glance

Related