Each agent uses one model, and different agents can use different ones. A model on your computer is free and works offline; a Localoy model runs in the cloud through the Localoy gateway and draws from your Localoy credit.
A free model on your computer
- Open Models. The first time, Get Qwen3 4B (2.5 GB) is a good start.
- Or search open models (try qwen, llama or gemma), or paste a link to any GGUF file under Any other model and press Download.
- When it has downloaded, pick it for an agent under On this computer in the agent’s model choice.
Models marked chat only can talk but can’t reliably use tools, so they suit a talking agent rather than one with switches. Bigger models want more memory. Each model shows whether it’s running, with Start and Stop, and you can delete it from its ⋯ menu.
A Localoy model
Localoy’s models (DeepSeek, Kimi, Claude Haiku, MiniMax and more) run through the Localoy gateway, so there is no key to set up. Pick one while making an agent, or in its Settings; each shows its price. They draw from your Localoy credit. In an organization, its admins choose which ones the team can use, and the spend goes to the team’s balance.
Keep spending in check
For a Localoy model, an agent’s settings have a daily limit in dollars. When the day’s spend reaches it, the agent stops until midnight; the chat offers Raise limit. The chat also shows what it has spent so far.