Skip to content

Tool Selection Guide

Loop Engineering is a runtime layer, not a foundation model or coding-agent replacement. It is designed to wrap repeatable workflows with explicit lifecycle controls.

Choose by workflow

Need Good starting point
Interactive code exploration and implementation A coding agent such as Claude Code or Codex
Current web information and general assistance A web-connected assistant such as Grok
Repeatable workflow with explicit state and budgets Loop Engineering around a model/tool adapter
One-off task Use the underlying assistant directly
Recurring task that must resume after interruption Use a persistent workflow runtime

Capabilities and commercial products change quickly. Verify current provider documentation before choosing a tool; this guide intentionally avoids checkmark comparisons that would become stale or imply unsupported superiority.

What Loop Engineering adds

  • Validated runtime states and transitions
  • Token, cost, step, and retry budgets
  • Deterministic checks before model calls
  • Separate generator and evaluator configuration
  • Configurable human checkpoints
  • Recovery handlers
  • JSON and SQLite persistence
  • A CLI for scaffolding, validation, auditing, execution, and estimates

These controls reduce operational ambiguity; they are not a security or correctness guarantee. External side effects require an explicit adapter and the default runtime does not silently simulate them.

Hybrid workflow

  1. Develop and inspect the task with your preferred coding agent.
  2. Capture the stable workflow as a SKILL.md contract.
  3. Configure conservative budgets, gates, and checkpoints.
  4. Connect only the external adapters the workflow needs.
  5. Run in dry-run mode, inspect state, then enable writes deliberately.

Cost notes

loop-engine cost produces planning estimates from configured assumptions. Actual cost varies by provider, model, prompt, output, caching, and retries. For supported live calls, the provider adapter records provider-reported token usage and cost when the API supplies it.

See also