The future is intelligence trained on your world.

Yes. Qova installs locally and works through your existing coding-agent account. Your source code, working project files, and existing instructions stay intact.

Qova’s learning updates only the agent configuration it manages. You can see what Qova adds, review its changes, and undo individual learnings.

Questions 2 and 4 explain what it writes and how to undo a change.

No. Qova’s learning updates only the agent configuration it manages, leaving your source code and working project files alone.

For Claude Code, this includes its own sections in CLAUDE.md, Qova-owned files inside .claude/, and files in Claude Code’s local project-memory folder. It follows the same approach for Codex, including its own sections in AGENTS.md.

Your existing instructions and configuration stay intact - Qova never deletes or overrides them.

Qova runs on your machine and uses your existing model provider through your own account and subscription. Your code, project files, conversations, and the learnings Qova creates are never shared with Qova Labs.

We collect only your email for authentication and basic usage and performance telemetry to understand how Qova is working for you.

You can undo individual learnings. Qova keeps a change history showing what changed and why. Use /qova undo with the learning’s ID from that history to remove it. Undo targets that learning while protecting unrelated content.

Qova’s learning runs in the background, so you can keep working without waiting for it to finish.

Memory stores information from past work. Qova is the learning layer above it: it uses your conversations, corrections, acceptance, decisions, and outcomes to improve how your agent approaches future work.

It continuously turns those lessons into focused instructions, skills, and workflows, then revisits them as your standards and project evolve, handling the ongoing harness maintenance that the best developers spend hours managing themselves.

For our fellow geeks out there - think of our work as skewed heavily towards meta-cognition and plasticity in neuroscience.

Compare real tasks, including your most difficult ones, from your own repo with Qova’s changes enabled and disabled. Keep the agent, model, and success criteria consistent. Look at how often tasks succeed, how many turns and retries they take, and how many tokens they use.

Qova currently supports Claude Code and Codex. Support for additional agents is planned. If you’d like to request access for another agent, please send us a note to support@qovalabs.com.

Qova builds on your existing CLAUDE.md and keeps your instructions always in place. It adds its learnings in clearly marked sections and manages those additions against a 150-line budget to keep them focused and manageable.

You can see what it contributed and review those additions whenever you like.

Qova stops learning and making further updates. Based on how you uninstall, you can choose to either remove all applied learnings or keep the instructions, rules, skills and other harness artifacts it has added as ordinary files your agent can keep using. You can keep, edit, or remove those files yourself.

Qova learns from your coding-agent sessions, including your corrections, preferences, decisions and successful thinking patterns and approaches. It turns useful lessons into instructions, skills, and other agent configuration that carry into future sessions. As you keep working, Qova keeps refining what it has learned.

Email support@qovalabs.com for help with installation, troubleshooting, or questions about Qova.