Claude Code, Codex, Gemini, Qwen, Mistral, OpenCode, Kimi, Pi. Each runs in an isolated container on your infrastructure. A clanker defines how each agent runs: model, runtime, repo scope, and the prompt templates that shape every phase. Set up one for quick copy tweaks, another for backend repos. Need another model? Scaffold a new agent plugin with one command.

Infrastructure-as-code sets up everything in your AWS account: the platform, agent runtimes, user management, and permissions. No code leaves your network.

Describe a change that touches the API, the web app, and the design tokens. The agent plans and executes across every affected repository. Research, Planning, Execution: every phase ends in a human review gate. You approve the direction before the agent spends a token on the next step.

Launch agent sessions with a slash command, pick your project and mode, and the agent streams progress into the thread. Reply to guide it, approve changes with a click, no tab-switching.

The Model Context Protocol lets external AI tools like Claude Desktop, Cursor, and Copilot interact directly with your Viberglass projects. List tickets, trigger agent runs, review plans, and approve work. The important tasks can all be achieved from inside the AI tool you already use. One API token, no custom glue code.

Dispatch agents on a cron to update dependencies, apply security patches, or run any recurring task. Claws run on schedule, open PRs, and notify you when they land.

Take screenshots, select elements, record your screen, and annotate, all from the browser. The extension captures console logs, network errors, viewport size, and browser details automatically. File a ticket with full context without leaving the page.

Every ticket is a shared workspace. Comment on research, revise the plan, send the agent back with new direction. Multiple people weigh in. The agent completes a full run, you review the output, and guide the next one. No IDE required.