Dashboard|

Models

Choose a model per agent and tune how it runs.

An agent takes any model in the Cencori catalog. Switching provider is changing one string.

new Agent({ name: 'Fast', model: 'gpt-oss-120b' });
new Agent({ name: 'Careful', model: 'claude-sonnet-4-5' });
new Agent({ name: 'Cheap', model: 'gemini-2.5-flash' });

The default

Without model, agents use gpt-oss-120b on Cerebras (DEFAULT_MODEL): thousands of tokens a second and reliable tool calling, so agents with several tool rounds stay quick.

Model settings

 
new Agent({
    name: 'Analyst',
    model: 'gpt-oss-120b',
    modelSettings: {
        temperature: 0.2,
        maxTokens: 2000,
        reasoningEffort: 'low', // 'none' | 'low' | 'medium' | 'high'
    },
});
SettingEffect
temperatureLower is steadier, higher more varied
maxTokensCap on each reply
reasoningEffortHow long a reasoning model thinks first. Lower is faster

Tools need a model that calls tools

Agents with tools, handoffs or widgets need a model that supports tool calling. Check capabilities.tools in GET /v1/models.

Your own keys

Bring your own provider keys (BYOK) in project settings and agents use them automatically: same code, your account's rates and limits.

Did you like the content?