An operator that isn’t tied to one AI company
OperatorNest is a model-agnostic AI agent: it runs on GPT, Claude, Gemini, Grok, Llama, DeepSeek, Mistral or Qwen, and you decide which one handles a given task. Each task runs on a model you can name, with no blended model behind the scenes, and you don’t need to be an expert in AI models to choose. Pick task by task, or leave it on a default. If a provider raises prices, changes its terms, or is a poor fit for a task, you switch, and your operator keeps working.
No single model is best at everything, and models, prices and terms keep changing. An operator built around one vendor’s model is only as good, as available and as fairly priced as that one company decides. See the Meta Muse vs Grok Bot comparison for two provider-tied assistants.
Choosing a model for a task
When you set up a task, or an operator dedicated to a kind of work, you pick the model it runs on. For example, run your research operator on a model that’s strong at synthesizing long documents, and your inbox operator on a faster model suited to short, frequent replies. You can use the AI model picker to compare model classes against a task’s needs, then change either choice at any time without starting over.
If you’d rather not think about it, leave it on Auto.
What Auto routing does
Auto looks at what a task needs and routes it to a model suited to that work: a stronger model for a research brief with a lot of comparison and judgment, and a lighter, faster one for a quick scheduling check. You can override it on any task.
Switching models without losing anything
Switching models on OperatorNest doesn’t mean starting over. Your operator’s memory, the task’s history, and anything you’ve corrected or taught it live with the operator and the task, not with the model that ran it. Move a task to another model partway through and it picks up with full context. See memory for what that context includes, or read how to switch from ChatGPT to Claude when you are moving work between those assistants.
Some agents on the market are built on a single company’s model with no option to change it. That’s a reasonable choice for a simpler assistant, and it ties you to that model’s strengths, limits and pricing for as long as you use the product. OperatorNest doesn’t ask you to make that trade-off.
Paying for the model you use
Model choice works alongside bring your own key: use a subscription or API key you already pay for with a provider, or use one of OperatorNest’s own. Either way, each task’s usage and cost are broken down by the model that ran it, so you can see what a research brief or a week of inbox triage cost in model usage.
A concrete example
Say your research operator runs on a model that’s strong at long, careful synthesis, and it’s midway through a twelve-source competitor brief when that provider raises its prices or has an outage. You switch that operator to a different provider. The brief continues from where it left off, using everything already gathered, drafted and remembered about how you like these briefs written. The only trace of the switch is a different model name in the receipt next to that step.
Where model choice shows up
Model choice is a setting on the task or the operator. It shows up in three places: when you first set up an operator, in a task’s options if you want to override the default for one job, and in that task’s receipt, which always names the model that produced the result. See receipts for the full record a task keeps.