Run coding agents against each other. Merge the winner.

Why? Because no single model is best for every task. We use selection because it leads to higher quality code.

Install via npm:

npm install -g voratiq

• Node 20+
• git
• 1+ AI coding agent (Claude >=2.0.55, Codex >=0.66.0, or Gemini >=0.19.4)
• macOS: ripgrep
• Linux (Debian/Ubuntu): bubblewrap, socat, ripgrep

See the sandbox runtime docs for guidance on the platform-specific dependencies.

Windows is not currently supported.

# Initialize workspace
voratiq init --yes

# Generate a spec
voratiq spec \
  --description "add dark mode toggle with localStorage persistence" \
  --agent <agent-id>

# Run agent ensemble against that spec
voratiq run --spec .voratiq/specs/add-dark-mode-toggle.md

# Review results
voratiq review --run <run-id> --agent <agent-id>

# Apply the best solution
voratiq apply --run <run-id> --agent <agent-id>

# Clean up workspace
voratiq prune --run <run-id>

For a detailed walkthrough, see the tutorial.

Voratiq positions you as the architect and reviewer, and shifts implementation onto an ensemble of agents.

In practice, the same spec goes to all agents, evals run automatically, and you pick the winner.

Every run (diffs, logs, eval results, and agent summaries) is local, configurable, inspectable, and fully auditable.

Learn about the Voratiq workflow and CLI:

• Tutorial - End-to-end walkthrough
• Core Concepts - Mental model and design philosophy
• CLI Reference - All commands and options

How to configure agents, evaluations, and execution environments:

• Agents - Define which agents run and how to invoke them
• Environment - Configure runtime environments
• Evals - Define checks that gate agent output
• Sandbox - Network and filesystem restrictions

Voratiq is available under the MIT License.