Your experienceIn every task
Less explaining, steering and correcting. More room for the work you want to do.
Why can AI do so much, yet still need so much from you?
Varyz learns from your past conversations and work, bringing your goals, decisions and standards into each new task. It gives your AI your vision, taste, and creative perspective, so you spend less time explaining, steering and correcting.
Less explaining: bring in the context
Before work starts, Varyz tells the agent how you operate, the project's direction and what you have already settled. The briefing covers the task at hand. In Claude Code and Codex a hook delivers it with your first prompt; in Cursor the agent asks for it.
When a task depends on earlier work, the agent can retrieve relevant conversations, connected project documents and Git history, with sources and dates. Recorded session reports help it see what an earlier session finished or left open. Those reports are the agents' accounts of their work, rather than independent verification that the work was completed.
Less steering: use the decisions you already made
Who is this for? What tone should it use? What should it leave out? When a choice depends on you, the agent asks Varyz through the Model Context Protocol (MCP). Varyz looks up what you already decided, keeping general preferences and project choices in their proper scope. Your agent still makes the routine implementation decisions you entrusted to it.
Fewer interruptions: group the new questions
When a necessary preference or fact is missing, questions come together as short choices you can answer quickly. Your reply can answer the next matching question, instead of being left behind in one chat. The agent should ask rather than guess when Varyz cannot establish your answer.
In away mode, questions wait for your return. Away mode never grants extra permission to publish, spend or share data.
Less to correct: carry your standards through to the result
Before the agent reports done, Varyz gives it your checklist for that kind of work. The agent goes through the requirements and tells you what remains. You review the result with your standards already part of the task.
Varyz supplies context and guidance; the agent plans, builds and checks. It cannot guarantee correct output, catch every mistake or ensure that every host invokes its tools. You still review the work and approve actions that need your permission.
How it learns
Varyz learns how you work from your past messages to AI agents. It reads eligible conversations through your connected agent, on your own plan, looking for each project's aim, your taste, how you like work done and what you already decided. Anything sensitive is skipped, and passwords and keys are removed.
- Strong evidence becomes a trial rule. When you corrected an agent over something, or said it in two conversations or projects, the rule takes effect on trial, and agents are told you have not confirmed it yet.
- The rest waits for your review. You keep or drop what Varyz found.
- New instructions count from now on. When you tell an agent "from now on" or "always", Varyz picks it up for your next task.
Habits and one-offs
Some instructions describe how you always work. Others belong to one project or one task. "Every website gets a privacy page" applies to every project; "this project's accent colour is violet" stays with one project. A task answer stays with that task. Reusable rules can be saved for one project or for all your projects.
What stays yours
- It runs on your computer. No account and no API key. What you teach it is stored on your disk.
- You stay in charge of it. Answers and rules are kept as a history. Review, replace or take back any of them, and agents hear the new version from then on.
- Cloud is optional. Varyz Cloud brings selected preferences and separately approved project history to connected AI tools. Local Varyz works without it.
Read how Varyz compares with CLAUDE.md, AGENTS.md and Cursor rules, or see how it is measured. Download local Varyz for free.
Start with what you already know.
Your preferences, your project history and your standards, ready for the next task.