TOXCI AGENT
ONE PROMPT. A COMPLETE PROJECT.
Login

YOUR ENGINE. YOUR CONTROL.

Runner & settings

Connect real execution infrastructure for native tests, builds, and deployment.

Workspace preferences

Appearance, editing and agent behavior — applied to the real workspace.

Saved on this browser. Sign in to save a separate set of private account settings. These preferences never bypass execution or deployment approval.

Used for empty projects; existing projects keep their technology.

Applies to generation smoke tests only. Native command reruns still require separate approval.

Wrapped lines hide the line-number gutter to avoid misleading alignment.

Added to future generation and repair requests. Never include passwords or API keys.

AI chat & personalization

ChatGPT-style controls for TOXCI's built-in AI. No ChatGPT subscription, model selector, voice service or live web search is connected. Explanation settings never shorten required source files or change approval rules.

Choose a language or type any language or dialect. Auto matches your prompt. This changes AI replies, not application labels.

When personalization is enabled, this profile and your agent instructions are sent with future AI requests. Do not include secrets or sensitive personal information. This is explicit saved context, not automatic learned memory.

Privacy & data controls

Conversation context controls what prior messages are sent to AI, not whether project history is saved. Project files and chat remain in your existing project storage. Export or delete individual projects from Projects. Reset the settings draft and Save to clear your saved personalization. Turning off personalization does not erase the saved profile. These controls do not change the AI provider's retention policies.

Import reference settings

Reference: sahid-ai-studio-remix.genmb.com. The linked app's settings could not be read automatically. Upload screenshots of its settings tabs or paste their labels and selected values. Only supported preferences above can be mapped; integrations and credentials are not copied.

No screenshot selected. Remove all visible secrets first. Images are analyzed, not saved as project files.

Execution runner

GenMB functions → dedicated HTTPS Docker host

Backend-only credentials

Set TOXCI_DOCKER_URL and TOXCI_DOCKER_TOKEN in GenMB backend Secrets. Use a dedicated, authenticated HTTPS proxy to the Docker Engine API. The URL and token are never supplied by visitors or sent to the AI.

Sign in with the app owner account before checking the runner, executing builds, or deploying. Every privileged operation verifies ownership on the server.

Configure the execution host

# Install reviewed images on your dedicated Docker host
docker pull node:22-bookworm-slim
docker pull python:3.12-slim
# Backend Secrets: TOXCI_DOCKER_URL, TOXCI_DOCKER_TOKEN

Backend endpoints run as GenMB Python functions, not a Node HTTP server. Docker runs only on your dedicated host; no Docker daemon is assumed in GenMB. Keep its proxy behind a firewall or private tunnel with authentication. Functions use fixed build recipes, return actual exit codes and artifacts, and clean up containers. Optional Netlify deployment uses backend-only NETLIFY_TOKEN and NETLIFY_SITE_ID.

Connected capabilities

AI generation

Real GenMB AI, no model API key needed. Responses are AI-generated, not live web research.

Realtime activity

Connected. Real SDK events report workspace activity without publishing project source or credentials.

Project persistence

Local files and history use IndexedDB. Sign in on the Account page, then use Save to cloud and Restore cloud projects for private cross-device saving in the built-in key-value store. A managed relational database is a Pro feature; you can also connect your own Supabase project free from the editor Database panel.

Safe by default

  • Opaque browser sandbox with network blocked.
  • Owner-checked GenMB functions, backend-only proxy token, disposable containers.
  • Source review and approval before every native execution.
  • No service credentials or signing keys in project files.
  • Production deployment and possible charges require explicit approval.

This is a single-operator MVP, not a multi-tenant security boundary. Use a dedicated runner host and a rootless Docker engine.

⚡Make an app like this, free