Self-hosted AI that knows your code, your docs and your designs.

Every answer cites the file it came from. Open-weight models run on your own GPU server: your code never leaves your infrastructure.

Knowing how your own system works takes too long, and cloud AI is not an option for your code.

export function applyDiscount(cart, code) {
  const rule = findRule(code);
  if (!rule || rule.expiresAt < now())
    return reject("invalid");
  if (cart.subtotal < rule.minBasket)
    return reject("min_basket");
  const amount = rule.type === "percent"
    ? Math.min(cart.subtotal * rule.value, rule.cap)
    : rule.value;
  // applied before tax
  return { ...cart, discount: amount };
}

Illustrative example. The codebase, files and design are invented.

Connect, index, ask, all on your servers.

1. 1

   Connect

   Point isobata at your GitLab projects (cloud or self-managed), your Figma files and the Markdown docs in your repositories.

2. 2

   Index

   It reads the code, documentation and designs on your own server.

3. 3

   Ask

   Your team asks in a chat and gets answers that list the files and frames they come from.

• Self-hostedInstalled in your cloud (VPC) or on-premises, on a single GPU server you control.
• Open-weight models on your own GPUNo third-party AI provider and no external AI API. The models run next to your code.
• Never trained on your codeWe never see your code, your questions or the answers.
• Outbound calls only to hosts you allowYou set the allowlist, and every outbound call is logged.
• A Docker image, security checks already passedisobata ships as a Docker image you run in your VPC. Every release is signed and vulnerability-scanned before we publish it, and verified again before it installs.
• Nothing reports backThe licence is checked offline on your server. No telemetry.

Questions