What if your AI had your guidance without waiting for your next prompt?Varyz draws on your past decisions and corrections to answer settled questions, guide the work and check it against your standards.
Varyz learns recurring preferences from eligible AI conversations and brings them into future tasks. Spend less time repeating what matters to you. Review and change what it learns whenever you need to.
Built-in memory learns after you repeat yourself. Varyz learns before.
Your local projects, their changes and the conversations behind them give your next agent a place to start. Find what you tried, why you changed direction and where recorded work left off, with sources you can check.
A normal AI starts from zero. Yours starts with the story.
When your agent needs a decision you've already made, Varyz can answer from that history. Necessary new choices wait for you while the agent continues the work it can do.
Nearly three times fewer interruptions while you step away.
Away mode keeps the work moving within existing permission. Publishing, spending and sharing still need your approval.
Your requirements become a checklist the agent uses before reporting finished work. The phone layout, the privacy page, the detail you always check. Your expectations are already in place when you review the result.
More than twice the standards met, before you even look.
Your agent still does the work, and you still review it. A checklist helps it check your requirements; it cannot guarantee a correct result.
Local context stays with you. No Varyz account required. Available offline once set up.
Cloud and the Varyz network are optional. Nothing is shared until you approve it.
Review, replace or withdraw learned preferences. Sensitive content is withheld.
Your AI tool may have its own network and account requirements. Public source release is being prepared under AGPL-3.0.
Keep your context local. Use Cloud across your tools, share project preferences with collaborators and follow creators' published standards.
Your preferences, decisions, relevant local history and finish checklists. No Varyz account required.
Download availabilityPublic release is being prepared. Downloads are not available yet.
Or $108 per year.
And much more
Explore CloudPlanned pricing in US dollars, tax included. Public paid signup is not open yet. History rollout is still being verified.
Shared and creator features are in a limited pilot. Following creators is free with a Cloud account; publishing needs the paid plan.
Varyz gives connected AI tools access to your preferences, goals, past decisions and relevant work history. It learns from eligible AI conversations, briefs agents for new tasks, answers from decisions you've already made and gives them your checklist before they report done.
No. Varyz, spelled with a z, provides personal context and preferences for AI agents at varyzai.com. Varys AI is a separate interior-design product.
Your agent still plans and builds. Varyz supplies the personal and project context that helps it do the work your way, using the AI tools you already work with.
Yes. Varyz helps the agent use your context and check your requirements; it cannot guarantee correct output or catch every mistake. You still review the result and approve actions that need your permission.
The local release supports Claude Code, Codex and Cursor. Guided installation is available from the download page. Any agent that supports the Model Context Protocol (MCP) can use it too.
Built-in memory and files such as CLAUDE.md, AGENTS.md or Cursor rules already help agents use context. Varyz combines learning from eligible conversations, retrieving relevant history, reusing scoped decisions and delivering finish checklists across connected tools. You can keep your instruction files for shared project conventions. See the comparison.
Routine implementation choices stay with your agent. Varyz helps it reuse settled decisions and group necessary new questions. Away mode defers questions until you return; it does not grant extra permission to publish, spend or share data.
It's stored on your own computer, and Varyz needs no account. Learning uses eligible past AI conversations, with sensitive content withheld. Agents receive relevant context for their task. Cloud sharing is optional: select confirmed preferences and approve project history separately. Sharing anonymized examples with the Varyz network is a separate opt-in that you can turn off anytime.
Varyz is open source under the AGPL-3.0 licence. Find the source code and installation instructions on the download page.
The planned local release will be free. Optional Varyz Cloud is preparing for launch at $12 a month or $108 a year, tax included, with no free trial and a 14-day refund window after each payment. You do not need cloud to use local Varyz. See cloud features and limits.
Yes. Answers and rules are kept as a history, so you can take one back or replace it, and agents hear the new version from then on.
Varyz Teams is planned: shared company decisions alongside each person's private preferences, with new questions routed to the responsible person. Team accounts and shared workspaces are not available yet. Today, keep shared conventions in your project's instruction files and use Varyz for your own preferences.
Varyz Teams is planned: shared company decisions alongside each person's private preferences. New questions would reach the responsible person, with a common standard for the work.
Team accounts and shared workspaces are not available yet. Today, use project instruction files for shared conventions.
Talk about TeamsInstall free. It learns from the conversations you’ve already had.
Get freeThe free local release is being prepared. Check download availability and setup details.