Liveform

Liveform learns what you like and catches what you miss

It works beside your coding agent in Claude Code and Cursor. It picks up your taste from what you ask for, checks what the agent builds, and speaks only when something matters.

An example session 0 in .liveform
    Or say it yourself

    Your agent keeps forgetting...

    Coding agents are quick and capable. Each session still starts without three things.

    01

    Personal taste

    You fix the spacing on one page, then ask again on the next. Every repeat costs time and tokens.

    02

    Agentic decisions

    Agents make app-wide choices you never asked for, such as how sign-in works.

    03

    Rigorous standards

    Consent before tracking. Labels a screen reader can use. Keys kept out of the repo. A fast build skips them without telling you.

    Keep building. Liveform remembers.

    It works beside your coding agent in Claude Code and Cursor. It picks up your taste from what you ask for, checks what the agent builds, and speaks only when something matters.

    Learns from what you ask

    Say “make it cleaner” and Liveform records what that means. There is no confirmation step.

    Checks what your agent builds

    When the agent stops, Liveform checks the change. A clear fix goes back as one instruction.

    Outputs in plain English

    Each project has one .liveform: your taste and the rules, in plain English.

    Quiet until it matters

    Liveform stays out of your session. It tells you directly for two reasons only: a risk, or a decision you should own. Everything else it fixes quietly or leaves to you.

    Tells you

    Risk

    Harm to your users, exposed access, or a law that applies.

    Tells you

    Ownership

    An app-wide decision the agent made, recorded so you can explain it later.

    Fixes quietly

    Fixes

    No tracking before consent, no personal data in logs, accessible form labels, a way to delete a user’s data.

    Never decides

    Left to you

    Your lawful basis, how long you keep data, what your privacy policy says, whether a use of AI is high-risk.

    Checked against named standards

    Law, accessibility and security practice. Each check says which rule it comes from, so you can see what it is based on.

    Apply legal standards without asking

    GDPR, the ePrivacy Directive, the European Accessibility Act and Article 50 of the AI Act. Ireland and the EU at launch.

    Liveform checks. It does not certify, and it will never tell you a product is compliant.

    Accessibility practice

    WCAG 2.2 AA.

    The standard that accessibility law points to. It covers the things a fast build forgets: labels a screen reader can use, text you can read against its background, and controls that work from a keyboard.

    Security practice

    The OWASP Top 10.

    The ten most common ways web apps are broken into. It covers the mistakes an agent makes without telling you: keys left in the repo, pages anyone can reach, and input that is trusted when it should not be.

    Secure by default

    Prompts and changes are screened on your machine before anything is sent.

    Files

    Secret files stay put

    The contents of .env files, key files and credential stores are never sent. They stay on this machine.

    Data

    Sensitive data obfuscated

    Keys, emails, phone numbers and card numbers leave as a label. sk_live_51H leaves as “API key”.

    Code

    Your code is never stored

    Liveform keeps your taste and your project rules, screened for secrets.

    It never writes to your code. Fixes go to your agent, and the agent makes the change.

    Try it on a project. For Claude Code and Cursor.

    By signing up, you agree to our Privacy policy

    Questions

    Does Liveform change my code?

    No. Liveform never writes to your code. When something needs fixing, it sends your agent a plain instruction that names the file and the change, and the agent does the work.

    Will it slow my agent down?

    No. Learning runs in the background while you prompt. Checks run when the agent stops, and anything that takes too long waits for the next turn.

    Will it keep interrupting me?

    No. You see one line when Liveform asks the agent to fix something. It tells you directly only for a risk, or for an app-wide decision you did not ask for. It never raises the same thing twice.