Choosing a model
Choose between managed cloud and local models, understand credit multipliers, and match tool support and reasoning effort to the task.
The model picker sits in the composer. It separates managed cloud models from supported local models and shows the capabilities relevant to the active mode.
Cloud and local routes
Cloud models are administered through Kosmico. The picker shows the models available to your account, their tool support, reasoning controls, and credit multiplier. The Recommended choice follows the managed default rather than saving a hidden model name.
Local models run through Ollama on the desktop device. Their prompts and selected project context stay on that device. Local inference does not make a cloud-synced project local; file synchronization and model routing are separate choices.
The available catalogue can change. Use the picker and the live help answer for current models rather than relying on a static list in documentation.
Credit multipliers
Cloud models can consume the included allowance at different rates. A 1× model uses the baseline rate, while a higher multiplier uses the allowance proportionally faster. A model marked free does not consume that allowance. The picker shows the applicable value and remaining usage.
Use a capable tool-supporting model for a long agent run where mistakes would create more work. Reading, summarizing, and formatting often need less. Kosmico preserves an explicit model selection for that conversation; Recommended follows the current policy.
Modes and reasoning
Agent and Plan require a model that supports tools. Chat-only models remain available for conversation but cannot execute those modes. The mode control explains the boundary for the selected model.
Reasoning effort is a separate preference. Models expose only the levels they support, from immediate answers to deeper reasoning. More reasoning can take longer and use more credits.
Run locally
Open Settings → Connections → Local AI in the desktop app. Kosmico checks CPU, memory, GPU, and free storage, then labels compatible models by fit. Download a recommended model or select a model already available through Ollama.
Local AI and the Connections screen are desktop-only. The web app uses the managed cloud route.