Models and task levels
How Nelix Agent matches the size of the model to the size of the task.
Nelix Agent is built on large language models. It does not use one model for everything. Each kind of task has a level, and each level is served by a model suited to it.
The three levels#
| Level | Used for | Why |
|---|---|---|
| Fast | Explaining a failed stage from its log | A short, bounded reading. Speed matters more than depth |
| Standard | Design Chat, and generating a testbench | Writing and explaining code, where a quick answer keeps you working |
| Deep | Diagnosis and repair, and timing and implementation advice | Reasoning across a design, a waveform and measured results |
Which model serves each level is set by Nelix and can change as better models appear. The level a task belongs to is fixed by the kind of task, not chosen per request.
What does not change with the model#
The rules an agent follows, the documents it is given, the checks on its answer and the limits on what it may do are the same whichever model answers. A stronger model is not trusted more.
What you will notice#
- A stage explanation appears quickly.
- A Design Chat reply is written out as it arrives.
- A diagnosis takes longer, because it reads more and reasons further.
Availability#
nelix ai status shows whether Nelix Agent is available. The three stages never depend on it: nelix test, nelix synth and nelix build run whether or not the AI is reachable.
Something unclear or missing? Send feedbackSign in and use Feedback to tell us.