NateBJones-Projects/ringer

What it solves

Ringer addresses the high cost and unreliability of using expensive frontier AI models for routine implementation tasks. It allows you to delegate complex planning and review to a premium model while offloading the actual coding, migrations, and testing to a swarm of cheaper, parallel workers.

How it works

You define a task manifest (a JSON file) that includes a specification for the AI, a shell command to verify the output, and expected files. Ringer spawns multiple workers in parallel to execute these tasks. Crucially, it does not trust the worker's claim of completion; it automatically executes your provided check command. If the check fails, Ringer retries the task once with the failure context injected. All activity is tracked in a local, real-time dashboard called Ringside.

Who it’s for

Developers and teams who want to scale their AI-assisted development workflows without scaling their costs linearly with every line of code. It is ideal for those who need to automate repetitive tasks like scaffolding, batch transforms, or test-suite generation while maintaining strict verification of the results.

Highlights

• Parallel Execution: Runs multiple worker agents simultaneously to speed up build and implementation tasks.
• Verification-First: Uses executable shell commands to verify results, ensuring that "done" actually means working code.
• Cost-Efficient: Enables the use of cheaper, specialized worker models for mechanical tasks while reserving premium models for high-level decision-making.
• Ringside Dashboard: Provides a local, real-time web interface to monitor swarm progress, token usage, and task verdicts.
• Pluggable Engines: Supports multiple worker CLIs (like Codex, Grok, or OpenCode) through a flexible configuration system.
• Safety & Control: Includes features like task-level sandboxing, sensitive data redaction, and a linting tool to catch common swarm configuration errors.