> ## Documentation Index
> Fetch the complete documentation index at: https://docs.workser.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# AI models: choose what an AI agent thinks with

> Pick a model for one chat, set an AI agent's default and fallback, understand Auto, Fast and Smart, and see which plans can choose.

**Every paid plan can choose the AI model for each AI agent and for each chat.** A chat choice wins for that Task; an AI
agent's default is what new chats on Auto and its Automations use. Running Tasks keep the model they started with.

## Pick a model for one chat

The model chip above the message box opens the picker. With nothing chosen it reads **Auto**; otherwise it shows the model,
or **Fast** / **Smart**.

* **Auto** lets Workser choose: **Fast** is "Quick answers at the lowest cost. Good for simple Tasks." **Smart** is "Deeper
  thinking on every step. Uses more credits."
* Filter by **work** (Design, Slides, Video, Coding, Writing, Research and more), by **tier** (Quick, Standard, Advanced), or
  search by name.
* Each row shows the provider and its price; the count at the foot says how many models match.

The picker's own note is the rule: "Remembered across chats. Automations use each agent's default. Running tasks keep their
model."

If you pick a model while chatting with one AI agent, Workser offers **Set \<name>'s default model?** — **Skip** keeps your
chat selection only; **Save for this agent** makes it the default for new chats on Auto and for its Automations.

## Set an AI agent's default

**Panel → Profile → AI model** is the default for that AI agent. It saves immediately ("Saved for new chats and
automations.") and its footnote repeats the priority: "Used for new chats on Auto and automations. A model selected in chat
takes priority for that task."

The folded **Model results** shows only recorded evidence from the last 30 days: average recorded cost, how often the final
check said done, customer feedback, extra check attempts and average elapsed time — then either a suggestion or "Not enough
comparable evidence to suggest a model change yet."

## A fallback and fine tuning

**Manage → Instructions & model** has the model fields:

* **Primary model**: what the AI agent thinks with.
* **Use a fallback model**: used when the main model is unavailable. A fallback cannot be set without a primary.
* **Advanced** per model: **Creativity** (temperature, 0–2: lower is steadier, higher is more varied) and **Max response
  length** (tokens, the longest single reply).

## Which plans can choose

Starter, Pro and Team choose premium models. A business without a plan runs one fast model and cannot pick: every picker is
replaced by **"Choose a plan to pick better AI models."** See [Plans and credits](https://docs.workser.ai/ai-workforce/plans-and-credits.md).

## Related

* [Chats](https://docs.workser.ai/using/chats.md): where the per-chat picker sits
* [The AI agent panel](https://docs.workser.ai/using/agent-panel.md): Profile and the Manage dialog
* [Autonomy and probation](https://docs.workser.ai/using/autonomy-and-probation.md): the other half of "how it works"


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.