AI routing should be accountable, reversible, cost-aware, and controlled by the operator.
We believe useful AI is not the model that speaks the loudest. It is the system that chooses carefully, explains its limits, protects what it touches, and stays answerable to the people who run the business.
ToggleLogic exists because a default model is not a strategy. Businesses deserve a practical routing layer that makes deliberate choices visible, preserves operator control, and improves only when evidence supports the change.
People set the business objective and retain the right to choose. Explicit owner and user selections take precedence over automatic routing. A system that cannot be overridden is not an assistant; it is an unaccountable dependency.
The point is not to use the cheapest model for everything. The point is to use the least expensive capable model for the work at hand, while retaining a clear path to stronger capability when the task truly requires it.
A silent downgrade creates false confidence and surprise costs. When a requested route, model family, or advanced layer is unavailable, the system should say so in its records and fail safely to the host's normal behavior rather than pretend the requested policy ran.
Routing should not require a vendor to collect customer prompts, business records, credentials, or local operating history. ToggleLogic Free keeps its public cost and routing records local. Advanced capabilities must earn trust without turning private work into somebody else's training data.
A model choice that worked once is an observation, not a universal rule. The Intelligence product is being built around verified outcomes, human approval, narrow scope, reversibility, and privacy-safe aggregation. It must never promote itself on the basis of raw chats or model self-congratulation.
The Free plugin is not a teaser for hidden magic. The Intelligence engine is also public for inspection and contribution. The production registry, maintained benchmark intelligence, private evidence, credentials, signed packs, and services remain separate because a public engine and a maintained commercial intelligence product serve different purposes. The boundary is deliberate, documented, and enforced.
Version numbers, tests, compatibility checks, release notes, and rollback paths matter. We will not hide a behavior change behind marketing language. When a release fixes a problem, the changelog should say what changed, what did not, and how it was verified.
Use the Free plugin. Read the release notes. Test it against real work. Tell us where the contract is unclear and where the controls should be stronger. Support means helping make AI orchestration more disciplined, more useful, and more accountable.
Public Product Edition 1.1.0 · Aligned with the Intelligence public-development policy. This page states product direction. It does not disclose proprietary registry data, customer information, credentials, private evidence, or legal records.