DistilCodeAll techniques

Technique 01 / 14

AutoJev model selectionPro

Match the starting model to the task instead of using the same class for every session.

Interactive model

See the change.

Three different tasks, each starting a new session. Compare a fixed model with AutoJev’s task-matched selection.

AutoJevtask-matched model

Fit the model to the work

  1. 01

    Rebuild the project

    Assigned model GPT 6.1 Luna
  2. 02

    Change UI make it look glass themed

    Assigned model GPT 6.1 Sol
  3. 03

    Change the provider from vercel to typesafe AI, follow docs for reference

    Assigned model GPT 6 Astra

Example model assignments. AutoJev uses the model configured for each selected class.

How it works

Less context. Same thread of work.

AutoJev sends up to 4,000 characters of the first normal prompt through the DistilCode API to the Jev provider (Vercel or typesafe AI), solely for model selection with zero data retention. No repository files are attached, and DistilCode does not store the prompt excerpt. Jev selects a model class, and DistilCode uses the configured model for that class before submitting the coding request directly to your chosen provider.

This can avoid using a more expensive model than a task needs. Selection happens only when AutoJev is enabled and a model is configured for the chosen class; savings depend on the task and selected model. Free includes 5 AutoJev selections; Pro includes 2,000 a month.

In practice

The operating loop

  1. 01

    Inspect the first prompt

    Up to 4,000 characters of the first normal prompt are sent to the AutoJev classifier.

  2. 02

    Choose a class

    Jev returns a model class suited to the requested work.

  3. 03

    Use the configured model

    DistilCode switches to the model assigned to that class before sending the prompt.