Models and AI tasks

One model runs every AI task unless you give a task its own, all chosen from OpenRouter's current list.

Every AI step runs on a model you choose, through your own OpenRouter key. You choose in Settings, AI. CareerBot has no built-in model: until you choose one, AI work can't run, and a step says "Choose a model for reading narratives in settings." (with its own task's name).

One model for every task

Under Model, choose the model every task uses, from reading your stories to writing letters. A few recommended models are listed, each with a line on what it's good at and what it would cost:

• Before CareerBot has done any work, the cost shows as OpenRouter's price per million tokens read, then written.
• After that, it shows what your last 30 days of AI work would cost a month with that model, at today's price.

To use a model that isn't listed, open Advanced and choose it under A model that isn’t listed. Any OpenRouter model works, including one that always points at a family's newest model (such as "Claude Haiku Latest"): each call goes to the model it points at then, at that model's price, and Compare models names the model that answered.

A task's own model

In Advanced, Tasks with their own model lists the tasks that don't follow the main model. Choose a model for a task gives one its own; Use the main model puts it back. Every other task keeps using the main model.

A task with its own model also has its own reasoning effort, for models that take one: the model's default, none, low, medium or high. Some models reason at length by default, which costs more.

Each step on The order names the task whose model it uses. These are the tasks:

• Reading narrativesRead a story, Read the changes, Check for disagreements, Read my goals, Filters from wording
• Reading projectsRead a project
• Rewriting factsRewrite a fact
• InsightsInsights
• Directions and search criteriaDirections and search criteria
• Screening companiesScreen companies, Fill in companies, Rank roles
• Follow-up questionsFollow-up questions
• Writing resumesResume, Write a resume line, Update lines
• Writing cover lettersCover letter, Update lines
• Ask about this roleAsk about this role
• Writing outreachOutreach
• Writing follow-upsFollow-ups
• Finding duplicatesFind duplicates
• Finding same workFind same work
• Gathering skillsGather skills
• Checking edited linesCheck against facts
• Test callsTest call
• Sorting rolesSort roles

Decision models

A decision model answers typed questions with a choice and a probability, never writing. OpenRouter offers several: Jev, Liquid's d1, Cloudflare's Clef, Respan's Span and others. Sorting roles is the one task that can take one: for each role and direction it answers whether the role's main work is that direction's work. Give Sorting roles its own model in Advanced and the list shows decision models beside chat models, marked Decision model, priced by what they read (their answers are free). They're never offered for any other task, nor as the main model. See Roles.

Try it

Try it sends one question to the model, to check your key and see how it answers. It shows the answer, the model, how long it took and what it cost, and is paid from your AI budget like any call.

Compare models

Compare models runs one task with up to four models side by side: reading a story, writing the base resume, or finding insights. Each run is paid from your AI budget; nothing it writes enters your record or resumes. When one model does best, Use this model makes it the main model, or that task's own.

Edit this page on GitHub