Why Viberglass exists.
Developers get pulled off feature work to fix button colors, update toast messages, and tweak copy. These tasks are easy to describe but require a full dev environment to implement. Viberglass gives the rest of your team a way to ship these changes through AI agents.
Why we built Viberglass
The pattern we kept seeing: someone reports a small issue, it sits in the backlog for weeks, or a developer drops what they're doing to fix it in 20 minutes. Neither outcome is good.
AI coding agents can handle these fixes, but they need orchestration. Something that manages the ticket, assigns the right agent, gates the work behind human review, and makes the result visible to the whole team.
We open-sourced Viberglass because infrastructure like this should be inspectable and self-hostable. The repo includes IaC to deploy to AWS, supports seven AI coding agents, and has user management built in. We're actively developing it and contributions are welcome.
What makes Viberglass different
Built for teams that deploy to AWS and review PRs.
Multi-Repository Support
Track and fix bugs across multiple GitHub repositories from a single ticket. One ticket can touch multiple repos.
Three-Phase Execution
Every ticket flows through Research, Planning, and Execution phases. Review and approve each stage before agents proceed.
Seven AI Agents
Claude Code, Qwen, Codex, Gemini, Mistral, OpenCode, and Kimi Code. Choose the right agent for each task at project or clanker level.
Ephemeral Workers
Agents spin up on demand in Docker, Lambda, or ECS. They run once and disappear, ensuring clean execution environments.
Clanker Configuration
Define runtime environments, agent instructions, and execution parameters with flexible configuration sets.
Inbound & Outbound Integrations
GitHub and Shortcut webhooks trigger automatic ticket creation. Agents create PRs that flow back to your workflow.
Tech stack
What Viberglass runs on.
Contribute to Viberglass.
The project is MIT licensed and actively developed.