Models, effort and speed

New model. Same Pom. Every bot has its own model, and you can change it whenever the job changes.

Pick a model per bot

Open the bot's model picker. Choose a connection, then a model. A good default team:

• Strategy and hard analysis: a top reasoning model such as Opus 5.5.
• Code and structured work: a strong coding model such as GPT-6.1.
• Chores, summaries, formatting: a fast, cheap model through OpenRouter.

The catalog stays current. When a new model ships and your connection can use it, it's there. Fluf lists what each connection can actually use; a model on the list in one connection doesn't mean every account has it.

Effort and speed

Effort sets how hard the model thinks. Speed picks a faster service mode where the provider offers one. Fluf shows each toggle only when your model and plan support it, so a missing toggle isn't an error.

Higher effort takes longer and uses more tokens. Faster modes can cost more or have different limits. Check your provider's terms.

Switch mid-conversation

Pick another model while you chat. The bot keeps its name, memory and full conversation history and carries on where it left off. If the bot is busy, let the current step finish or stop it; you can cancel a pending switch.

Each model still has its own context limit. The new model gets a handoff brief plus access to the saved history, so long conversations carry over without stuffing every old message into one prompt.

When a connection hits its limit

Fluf tells you which connection is getting full and which bot uses the most. Switch the bot to another model or connection, or resume later; its work is saved. Fluf only falls back to another account or paid key if you've allowed it.

Next: Connect your AI · Troubleshooting